drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2008 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #include <linux/module.h>
  30 #include <linux/types.h>
  31 #include <linux/init.h>
  32 #include <linux/pci.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/delay.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/tcp.h>
  38 #include <linux/ipv6.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44 #include <linux/cpu.h>
  45 #include <linux/smp.h>
  46 #include <linux/pm_qos_params.h>
  47
  48 #include "e1000.h"
  49
  50 #define DRV_VERSION "0.3.3.3-k2"
  51 char e1000e_driver_name[] = "e1000e";
  52 const char e1000e_driver_version[] = DRV_VERSION;
  53
  54 static const struct e1000_info *e1000_info_tbl[] = {
  55         [board_82571]           = &e1000_82571_info,
  56         [board_82572]           = &e1000_82572_info,
  57         [board_82573]           = &e1000_82573_info,
  58         [board_80003es2lan]     = &e1000_es2_info,
  59         [board_ich8lan]         = &e1000_ich8_info,
  60         [board_ich9lan]         = &e1000_ich9_info,
  61 };
  62
  63 #ifdef DEBUG
  64 /**
  65  * e1000_get_hw_dev_name - return device name string
  66  * used by hardware layer to print debugging information
  67  **/
  68 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
  69 {
  70         return hw->adapter->netdev->name;
  71 }
  72 #endif
  73
  74 /**
  75  * e1000_desc_unused - calculate if we have unused descriptors
  76  **/
  77 static int e1000_desc_unused(struct e1000_ring *ring)
  78 {
  79         if (ring->next_to_clean > ring->next_to_use)
  80                 return ring->next_to_clean - ring->next_to_use - 1;
  81
  82         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  83 }
  84
  85 /**
  86  * e1000_receive_skb - helper function to handle Rx indications
  87  * @adapter: board private structure
  88  * @status: descriptor status field as written by hardware
  89  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  90  * @skb: pointer to sk_buff to be indicated to stack
  91  **/
  92 static void e1000_receive_skb(struct e1000_adapter *adapter,
  93                               struct net_device *netdev,
  94                               struct sk_buff *skb,
  95                               u8 status, __le16 vlan)
  96 {
  97         skb->protocol = eth_type_trans(skb, netdev);
  98
  99         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 100                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
 101                                          le16_to_cpu(vlan) &
 102                                          E1000_RXD_SPC_VLAN_MASK);
 103         else
 104                 netif_receive_skb(skb);
 105
 106         netdev->last_rx = jiffies;
 107 }
 108
 109 /**
 110  * e1000_rx_checksum - Receive Checksum Offload for 82543
 111  * @adapter:     board private structure
 112  * @status_err:  receive descriptor status and error fields
 113  * @csum:       receive descriptor csum field
 114  * @sk_buff:     socket buffer with received data
 115  **/
 116 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 117                               u32 csum, struct sk_buff *skb)
 118 {
 119         u16 status = (u16)status_err;
 120         u8 errors = (u8)(status_err >> 24);
 121         skb->ip_summed = CHECKSUM_NONE;
 122
 123         /* Ignore Checksum bit is set */
 124         if (status & E1000_RXD_STAT_IXSM)
 125                 return;
 126         /* TCP/UDP checksum error bit is set */
 127         if (errors & E1000_RXD_ERR_TCPE) {
 128                 /* let the stack verify checksum errors */
 129                 adapter->hw_csum_err++;
 130                 return;
 131         }
 132
 133         /* TCP/UDP Checksum has not been calculated */
 134         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 135                 return;
 136
 137         /* It must be a TCP or UDP packet with a valid checksum */
 138         if (status & E1000_RXD_STAT_TCPCS) {
 139                 /* TCP checksum is good */
 140                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 141         } else {
 142                 /*
 143                  * IP fragment with UDP payload
 144                  * Hardware complements the payload checksum, so we undo it
 145                  * and then put the value in host order for further stack use.
 146                  */
 147                 __sum16 sum = (__force __sum16)htons(csum);
 148                 skb->csum = csum_unfold(~sum);
 149                 skb->ip_summed = CHECKSUM_COMPLETE;
 150         }
 151         adapter->hw_csum_good++;
 152 }
 153
 154 /**
 155  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 156  * @adapter: address of board private structure
 157  **/
 158 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 159                                    int cleaned_count)
 160 {
 161         struct net_device *netdev = adapter->netdev;
 162         struct pci_dev *pdev = adapter->pdev;
 163         struct e1000_ring *rx_ring = adapter->rx_ring;
 164         struct e1000_rx_desc *rx_desc;
 165         struct e1000_buffer *buffer_info;
 166         struct sk_buff *skb;
 167         unsigned int i;
 168         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
 169
 170         i = rx_ring->next_to_use;
 171         buffer_info = &rx_ring->buffer_info[i];
 172
 173         while (cleaned_count--) {
 174                 skb = buffer_info->skb;
 175                 if (skb) {
 176                         skb_trim(skb, 0);
 177                         goto map_skb;
 178                 }
 179
 180                 skb = netdev_alloc_skb(netdev, bufsz);
 181                 if (!skb) {
 182                         /* Better luck next round */
 183                         adapter->alloc_rx_buff_failed++;
 184                         break;
 185                 }
 186
 187                 /*
 188                  * Make buffer alignment 2 beyond a 16 byte boundary
 189                  * this will result in a 16 byte aligned IP header after
 190                  * the 14 byte MAC header is removed
 191                  */
 192                 skb_reserve(skb, NET_IP_ALIGN);
 193
 194                 buffer_info->skb = skb;
 195 map_skb:
 196                 buffer_info->dma = pci_map_single(pdev, skb->data,
 197                                                   adapter->rx_buffer_len,
 198                                                   PCI_DMA_FROMDEVICE);
 199                 if (pci_dma_mapping_error(buffer_info->dma)) {
 200                         dev_err(&pdev->dev, "RX DMA map failed\n");
 201                         adapter->rx_dma_failed++;
 202                         break;
 203                 }
 204
 205                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 206                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 207
 208                 i++;
 209                 if (i == rx_ring->count)
 210                         i = 0;
 211                 buffer_info = &rx_ring->buffer_info[i];
 212         }
 213
 214         if (rx_ring->next_to_use != i) {
 215                 rx_ring->next_to_use = i;
 216                 if (i-- == 0)
 217                         i = (rx_ring->count - 1);
 218
 219                 /*
 220                  * Force memory writes to complete before letting h/w
 221                  * know there are new descriptors to fetch.  (Only
 222                  * applicable for weak-ordered memory model archs,
 223                  * such as IA-64).
 224                  */
 225                 wmb();
 226                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 227         }
 228 }
 229
 230 /**
 231  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 232  * @adapter: address of board private structure
 233  **/
 234 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 235                                       int cleaned_count)
 236 {
 237         struct net_device *netdev = adapter->netdev;
 238         struct pci_dev *pdev = adapter->pdev;
 239         union e1000_rx_desc_packet_split *rx_desc;
 240         struct e1000_ring *rx_ring = adapter->rx_ring;
 241         struct e1000_buffer *buffer_info;
 242         struct e1000_ps_page *ps_page;
 243         struct sk_buff *skb;
 244         unsigned int i, j;
 245
 246         i = rx_ring->next_to_use;
 247         buffer_info = &rx_ring->buffer_info[i];
 248
 249         while (cleaned_count--) {
 250                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 251
 252                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 253                         ps_page = &buffer_info->ps_pages[j];
 254                         if (j >= adapter->rx_ps_pages) {
 255                                 /* all unused desc entries get hw null ptr */
 256                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 257                                 continue;
 258                         }
 259                         if (!ps_page->page) {
 260                                 ps_page->page = alloc_page(GFP_ATOMIC);
 261                                 if (!ps_page->page) {
 262                                         adapter->alloc_rx_buff_failed++;
 263                                         goto no_buffers;
 264                                 }
 265                                 ps_page->dma = pci_map_page(pdev,
 266                                                    ps_page->page,
 267                                                    0, PAGE_SIZE,
 268                                                    PCI_DMA_FROMDEVICE);
 269                                 if (pci_dma_mapping_error(ps_page->dma)) {
 270                                         dev_err(&adapter->pdev->dev,
 271                                           "RX DMA page map failed\n");
 272                                         adapter->rx_dma_failed++;
 273                                         goto no_buffers;
 274                                 }
 275                         }
 276                         /*
 277                          * Refresh the desc even if buffer_addrs
 278                          * didn't change because each write-back
 279                          * erases this info.
 280                          */
 281                         rx_desc->read.buffer_addr[j+1] =
 282                              cpu_to_le64(ps_page->dma);
 283                 }
 284
 285                 skb = netdev_alloc_skb(netdev,
 286                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
 287
 288                 if (!skb) {
 289                         adapter->alloc_rx_buff_failed++;
 290                         break;
 291                 }
 292
 293                 /*
 294                  * Make buffer alignment 2 beyond a 16 byte boundary
 295                  * this will result in a 16 byte aligned IP header after
 296                  * the 14 byte MAC header is removed
 297                  */
 298                 skb_reserve(skb, NET_IP_ALIGN);
 299
 300                 buffer_info->skb = skb;
 301                 buffer_info->dma = pci_map_single(pdev, skb->data,
 302                                                   adapter->rx_ps_bsize0,
 303                                                   PCI_DMA_FROMDEVICE);
 304                 if (pci_dma_mapping_error(buffer_info->dma)) {
 305                         dev_err(&pdev->dev, "RX DMA map failed\n");
 306                         adapter->rx_dma_failed++;
 307                         /* cleanup skb */
 308                         dev_kfree_skb_any(skb);
 309                         buffer_info->skb = NULL;
 310                         break;
 311                 }
 312
 313                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 314
 315                 i++;
 316                 if (i == rx_ring->count)
 317                         i = 0;
 318                 buffer_info = &rx_ring->buffer_info[i];
 319         }
 320
 321 no_buffers:
 322         if (rx_ring->next_to_use != i) {
 323                 rx_ring->next_to_use = i;
 324
 325                 if (!(i--))
 326                         i = (rx_ring->count - 1);
 327
 328                 /*
 329                  * Force memory writes to complete before letting h/w
 330                  * know there are new descriptors to fetch.  (Only
 331                  * applicable for weak-ordered memory model archs,
 332                  * such as IA-64).
 333                  */
 334                 wmb();
 335                 /*
 336                  * Hardware increments by 16 bytes, but packet split
 337                  * descriptors are 32 bytes...so we increment tail
 338                  * twice as much.
 339                  */
 340                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 341         }
 342 }
 343
 344 /**
 345  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 346  * @adapter: address of board private structure
 347  * @rx_ring: pointer to receive ring structure
 348  * @cleaned_count: number of buffers to allocate this pass
 349  **/
 350
 351 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 352                                          int cleaned_count)
 353 {
 354         struct net_device *netdev = adapter->netdev;
 355         struct pci_dev *pdev = adapter->pdev;
 356         struct e1000_rx_desc *rx_desc;
 357         struct e1000_ring *rx_ring = adapter->rx_ring;
 358         struct e1000_buffer *buffer_info;
 359         struct sk_buff *skb;
 360         unsigned int i;
 361         unsigned int bufsz = 256 -
 362                              16 /* for skb_reserve */ -
 363                              NET_IP_ALIGN;
 364
 365         i = rx_ring->next_to_use;
 366         buffer_info = &rx_ring->buffer_info[i];
 367
 368         while (cleaned_count--) {
 369                 skb = buffer_info->skb;
 370                 if (skb) {
 371                         skb_trim(skb, 0);
 372                         goto check_page;
 373                 }
 374
 375                 skb = netdev_alloc_skb(netdev, bufsz);
 376                 if (unlikely(!skb)) {
 377                         /* Better luck next round */
 378                         adapter->alloc_rx_buff_failed++;
 379                         break;
 380                 }
 381
 382                 /* Make buffer alignment 2 beyond a 16 byte boundary
 383                  * this will result in a 16 byte aligned IP header after
 384                  * the 14 byte MAC header is removed
 385                  */
 386                 skb_reserve(skb, NET_IP_ALIGN);
 387
 388                 buffer_info->skb = skb;
 389 check_page:
 390                 /* allocate a new page if necessary */
 391                 if (!buffer_info->page) {
 392                         buffer_info->page = alloc_page(GFP_ATOMIC);
 393                         if (unlikely(!buffer_info->page)) {
 394                                 adapter->alloc_rx_buff_failed++;
 395                                 break;
 396                         }
 397                 }
 398
 399                 if (!buffer_info->dma)
 400                         buffer_info->dma = pci_map_page(pdev,
 401                                                         buffer_info->page, 0,
 402                                                         PAGE_SIZE,
 403                                                         PCI_DMA_FROMDEVICE);
 404
 405                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 406                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 407
 408                 if (unlikely(++i == rx_ring->count))
 409                         i = 0;
 410                 buffer_info = &rx_ring->buffer_info[i];
 411         }
 412
 413         if (likely(rx_ring->next_to_use != i)) {
 414                 rx_ring->next_to_use = i;
 415                 if (unlikely(i-- == 0))
 416                         i = (rx_ring->count - 1);
 417
 418                 /* Force memory writes to complete before letting h/w
 419                  * know there are new descriptors to fetch.  (Only
 420                  * applicable for weak-ordered memory model archs,
 421                  * such as IA-64). */
 422                 wmb();
 423                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 424         }
 425 }
 426
 427 /**
 428  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 429  * @adapter: board private structure
 430  *
 431  * the return value indicates whether actual cleaning was done, there
 432  * is no guarantee that everything was cleaned
 433  **/
 434 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 435                                int *work_done, int work_to_do)
 436 {
 437         struct net_device *netdev = adapter->netdev;
 438         struct pci_dev *pdev = adapter->pdev;
 439         struct e1000_ring *rx_ring = adapter->rx_ring;
 440         struct e1000_rx_desc *rx_desc, *next_rxd;
 441         struct e1000_buffer *buffer_info, *next_buffer;
 442         u32 length;
 443         unsigned int i;
 444         int cleaned_count = 0;
 445         bool cleaned = 0;
 446         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 447
 448         i = rx_ring->next_to_clean;
 449         rx_desc = E1000_RX_DESC(*rx_ring, i);
 450         buffer_info = &rx_ring->buffer_info[i];
 451
 452         while (rx_desc->status & E1000_RXD_STAT_DD) {
 453                 struct sk_buff *skb;
 454                 u8 status;
 455
 456                 if (*work_done >= work_to_do)
 457                         break;
 458                 (*work_done)++;
 459
 460                 status = rx_desc->status;
 461                 skb = buffer_info->skb;
 462                 buffer_info->skb = NULL;
 463
 464                 prefetch(skb->data - NET_IP_ALIGN);
 465
 466                 i++;
 467                 if (i == rx_ring->count)
 468                         i = 0;
 469                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 470                 prefetch(next_rxd);
 471
 472                 next_buffer = &rx_ring->buffer_info[i];
 473
 474                 cleaned = 1;
 475                 cleaned_count++;
 476                 pci_unmap_single(pdev,
 477                                  buffer_info->dma,
 478                                  adapter->rx_buffer_len,
 479                                  PCI_DMA_FROMDEVICE);
 480                 buffer_info->dma = 0;
 481
 482                 length = le16_to_cpu(rx_desc->length);
 483
 484                 /* !EOP means multiple descriptors were used to store a single
 485                  * packet, also make sure the frame isn't just CRC only */
 486                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 487                         /* All receives must fit into a single buffer */
 488                         ndev_dbg(netdev, "%s: Receive packet consumed "
 489                                  "multiple buffers\n", netdev->name);
 490                         /* recycle */
 491                         buffer_info->skb = skb;
 492                         goto next_desc;
 493                 }
 494
 495                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 496                         /* recycle */
 497                         buffer_info->skb = skb;
 498                         goto next_desc;
 499                 }
 500
 501                 total_rx_bytes += length;
 502                 total_rx_packets++;
 503
 504                 /*
 505                  * code added for copybreak, this should improve
 506                  * performance for small packets with large amounts
 507                  * of reassembly being done in the stack
 508                  */
 509                 if (length < copybreak) {
 510                         struct sk_buff *new_skb =
 511                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
 512                         if (new_skb) {
 513                                 skb_reserve(new_skb, NET_IP_ALIGN);
 514                                 memcpy(new_skb->data - NET_IP_ALIGN,
 515                                        skb->data - NET_IP_ALIGN,
 516                                        length + NET_IP_ALIGN);
 517                                 /* save the skb in buffer_info as good */
 518                                 buffer_info->skb = skb;
 519                                 skb = new_skb;
 520                         }
 521                         /* else just continue with the old one */
 522                 }
 523                 /* end copybreak code */
 524                 skb_put(skb, length);
 525
 526                 /* Receive Checksum Offload */
 527                 e1000_rx_checksum(adapter,
 528                                   (u32)(status) |
 529                                   ((u32)(rx_desc->errors) << 24),
 530                                   le16_to_cpu(rx_desc->csum), skb);
 531
 532                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 533
 534 next_desc:
 535                 rx_desc->status = 0;
 536
 537                 /* return some buffers to hardware, one at a time is too slow */
 538                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 539                         adapter->alloc_rx_buf(adapter, cleaned_count);
 540                         cleaned_count = 0;
 541                 }
 542
 543                 /* use prefetched values */
 544                 rx_desc = next_rxd;
 545                 buffer_info = next_buffer;
 546         }
 547         rx_ring->next_to_clean = i;
 548
 549         cleaned_count = e1000_desc_unused(rx_ring);
 550         if (cleaned_count)
 551                 adapter->alloc_rx_buf(adapter, cleaned_count);
 552
 553         adapter->total_rx_bytes += total_rx_bytes;
 554         adapter->total_rx_packets += total_rx_packets;
 555         adapter->net_stats.rx_bytes += total_rx_bytes;
 556         adapter->net_stats.rx_packets += total_rx_packets;
 557         return cleaned;
 558 }
 559
 560 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 561                              struct e1000_buffer *buffer_info)
 562 {
 563         if (buffer_info->dma) {
 564                 pci_unmap_page(adapter->pdev, buffer_info->dma,
 565                                buffer_info->length, PCI_DMA_TODEVICE);
 566                 buffer_info->dma = 0;
 567         }
 568         if (buffer_info->skb) {
 569                 dev_kfree_skb_any(buffer_info->skb);
 570                 buffer_info->skb = NULL;
 571         }
 572 }
 573
 574 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 575 {
 576         struct e1000_ring *tx_ring = adapter->tx_ring;
 577         unsigned int i = tx_ring->next_to_clean;
 578         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 579         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 580         struct net_device *netdev = adapter->netdev;
 581
 582         /* detected Tx unit hang */
 583         ndev_err(netdev,
 584                  "Detected Tx Unit Hang:\n"
 585                  "  TDH                  <%x>\n"
 586                  "  TDT                  <%x>\n"
 587                  "  next_to_use          <%x>\n"
 588                  "  next_to_clean        <%x>\n"
 589                  "buffer_info[next_to_clean]:\n"
 590                  "  time_stamp           <%lx>\n"
 591                  "  next_to_watch        <%x>\n"
 592                  "  jiffies              <%lx>\n"
 593                  "  next_to_watch.status <%x>\n",
 594                  readl(adapter->hw.hw_addr + tx_ring->head),
 595                  readl(adapter->hw.hw_addr + tx_ring->tail),
 596                  tx_ring->next_to_use,
 597                  tx_ring->next_to_clean,
 598                  tx_ring->buffer_info[eop].time_stamp,
 599                  eop,
 600                  jiffies,
 601                  eop_desc->upper.fields.status);
 602 }
 603
 604 /**
 605  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 606  * @adapter: board private structure
 607  *
 608  * the return value indicates whether actual cleaning was done, there
 609  * is no guarantee that everything was cleaned
 610  **/
 611 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 612 {
 613         struct net_device *netdev = adapter->netdev;
 614         struct e1000_hw *hw = &adapter->hw;
 615         struct e1000_ring *tx_ring = adapter->tx_ring;
 616         struct e1000_tx_desc *tx_desc, *eop_desc;
 617         struct e1000_buffer *buffer_info;
 618         unsigned int i, eop;
 619         unsigned int count = 0;
 620         bool cleaned = 0;
 621         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 622
 623         i = tx_ring->next_to_clean;
 624         eop = tx_ring->buffer_info[i].next_to_watch;
 625         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 626
 627         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
 628                 for (cleaned = 0; !cleaned; ) {
 629                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 630                         buffer_info = &tx_ring->buffer_info[i];
 631                         cleaned = (i == eop);
 632
 633                         if (cleaned) {
 634                                 struct sk_buff *skb = buffer_info->skb;
 635                                 unsigned int segs, bytecount;
 636                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 637                                 /* multiply data chunks by size of headers */
 638                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 639                                             skb->len;
 640                                 total_tx_packets += segs;
 641                                 total_tx_bytes += bytecount;
 642                         }
 643
 644                         e1000_put_txbuf(adapter, buffer_info);
 645                         tx_desc->upper.data = 0;
 646
 647                         i++;
 648                         if (i == tx_ring->count)
 649                                 i = 0;
 650                 }
 651
 652                 eop = tx_ring->buffer_info[i].next_to_watch;
 653                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 654 #define E1000_TX_WEIGHT 64
 655                 /* weight of a sort for tx, to avoid endless transmit cleanup */
 656                 if (count++ == E1000_TX_WEIGHT)
 657                         break;
 658         }
 659
 660         tx_ring->next_to_clean = i;
 661
 662 #define TX_WAKE_THRESHOLD 32
 663         if (cleaned && netif_carrier_ok(netdev) &&
 664                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 665                 /* Make sure that anybody stopping the queue after this
 666                  * sees the new next_to_clean.
 667                  */
 668                 smp_mb();
 669
 670                 if (netif_queue_stopped(netdev) &&
 671                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 672                         netif_wake_queue(netdev);
 673                         ++adapter->restart_queue;
 674                 }
 675         }
 676
 677         if (adapter->detect_tx_hung) {
 678                 /*
 679                  * Detect a transmit hang in hardware, this serializes the
 680                  * check with the clearing of time_stamp and movement of i
 681                  */
 682                 adapter->detect_tx_hung = 0;
 683                 if (tx_ring->buffer_info[eop].dma &&
 684                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
 685                                + (adapter->tx_timeout_factor * HZ))
 686                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 687                         e1000_print_tx_hang(adapter);
 688                         netif_stop_queue(netdev);
 689                 }
 690         }
 691         adapter->total_tx_bytes += total_tx_bytes;
 692         adapter->total_tx_packets += total_tx_packets;
 693         adapter->net_stats.tx_bytes += total_tx_bytes;
 694         adapter->net_stats.tx_packets += total_tx_packets;
 695         return cleaned;
 696 }
 697
 698 /**
 699  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 700  * @adapter: board private structure
 701  *
 702  * the return value indicates whether actual cleaning was done, there
 703  * is no guarantee that everything was cleaned
 704  **/
 705 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 706                                   int *work_done, int work_to_do)
 707 {
 708         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 709         struct net_device *netdev = adapter->netdev;
 710         struct pci_dev *pdev = adapter->pdev;
 711         struct e1000_ring *rx_ring = adapter->rx_ring;
 712         struct e1000_buffer *buffer_info, *next_buffer;
 713         struct e1000_ps_page *ps_page;
 714         struct sk_buff *skb;
 715         unsigned int i, j;
 716         u32 length, staterr;
 717         int cleaned_count = 0;
 718         bool cleaned = 0;
 719         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 720
 721         i = rx_ring->next_to_clean;
 722         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 723         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 724         buffer_info = &rx_ring->buffer_info[i];
 725
 726         while (staterr & E1000_RXD_STAT_DD) {
 727                 if (*work_done >= work_to_do)
 728                         break;
 729                 (*work_done)++;
 730                 skb = buffer_info->skb;
 731
 732                 /* in the packet split case this is header only */
 733                 prefetch(skb->data - NET_IP_ALIGN);
 734
 735                 i++;
 736                 if (i == rx_ring->count)
 737                         i = 0;
 738                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 739                 prefetch(next_rxd);
 740
 741                 next_buffer = &rx_ring->buffer_info[i];
 742
 743                 cleaned = 1;
 744                 cleaned_count++;
 745                 pci_unmap_single(pdev, buffer_info->dma,
 746                                  adapter->rx_ps_bsize0,
 747                                  PCI_DMA_FROMDEVICE);
 748                 buffer_info->dma = 0;
 749
 750                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 751                         ndev_dbg(netdev, "%s: Packet Split buffers didn't pick "
 752                                  "up the full packet\n", netdev->name);
 753                         dev_kfree_skb_irq(skb);
 754                         goto next_desc;
 755                 }
 756
 757                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 758                         dev_kfree_skb_irq(skb);
 759                         goto next_desc;
 760                 }
 761
 762                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 763
 764                 if (!length) {
 765                         ndev_dbg(netdev, "%s: Last part of the packet spanning"
 766                                  " multiple descriptors\n", netdev->name);
 767                         dev_kfree_skb_irq(skb);
 768                         goto next_desc;
 769                 }
 770
 771                 /* Good Receive */
 772                 skb_put(skb, length);
 773
 774                 {
 775                 /*
 776                  * this looks ugly, but it seems compiler issues make it
 777                  * more efficient than reusing j
 778                  */
 779                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 780
 781                 /*
 782                  * page alloc/put takes too long and effects small packet
 783                  * throughput, so unsplit small packets and save the alloc/put
 784                  * only valid in softirq (napi) context to call kmap_*
 785                  */
 786                 if (l1 && (l1 <= copybreak) &&
 787                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 788                         u8 *vaddr;
 789
 790                         ps_page = &buffer_info->ps_pages[0];
 791
 792                         /*
 793                          * there is no documentation about how to call
 794                          * kmap_atomic, so we can't hold the mapping
 795                          * very long
 796                          */
 797                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 798                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 799                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 800                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 801                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 802                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 803                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 804
 805                         skb_put(skb, l1);
 806                         goto copydone;
 807                 } /* if */
 808                 }
 809
 810                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 811                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 812                         if (!length)
 813                                 break;
 814
 815                         ps_page = &buffer_info->ps_pages[j];
 816                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 817                                        PCI_DMA_FROMDEVICE);
 818                         ps_page->dma = 0;
 819                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 820                         ps_page->page = NULL;
 821                         skb->len += length;
 822                         skb->data_len += length;
 823                         skb->truesize += length;
 824                 }
 825
 826 copydone:
 827                 total_rx_bytes += skb->len;
 828                 total_rx_packets++;
 829
 830                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 831                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 832
 833                 if (rx_desc->wb.upper.header_status &
 834                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 835                         adapter->rx_hdr_split++;
 836
 837                 e1000_receive_skb(adapter, netdev, skb,
 838                                   staterr, rx_desc->wb.middle.vlan);
 839
 840 next_desc:
 841                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 842                 buffer_info->skb = NULL;
 843
 844                 /* return some buffers to hardware, one at a time is too slow */
 845                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 846                         adapter->alloc_rx_buf(adapter, cleaned_count);
 847                         cleaned_count = 0;
 848                 }
 849
 850                 /* use prefetched values */
 851                 rx_desc = next_rxd;
 852                 buffer_info = next_buffer;
 853
 854                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 855         }
 856         rx_ring->next_to_clean = i;
 857
 858         cleaned_count = e1000_desc_unused(rx_ring);
 859         if (cleaned_count)
 860                 adapter->alloc_rx_buf(adapter, cleaned_count);
 861
 862         adapter->total_rx_bytes += total_rx_bytes;
 863         adapter->total_rx_packets += total_rx_packets;
 864         adapter->net_stats.rx_bytes += total_rx_bytes;
 865         adapter->net_stats.rx_packets += total_rx_packets;
 866         return cleaned;
 867 }
 868
 869 /**
 870  * e1000_consume_page - helper function
 871  **/
 872 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 873                                u16 length)
 874 {
 875         bi->page = NULL;
 876         skb->len += length;
 877         skb->data_len += length;
 878         skb->truesize += length;
 879 }
 880
 881 /**
 882  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 883  * @adapter: board private structure
 884  *
 885  * the return value indicates whether actual cleaning was done, there
 886  * is no guarantee that everything was cleaned
 887  **/
 888
 889 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 890                                      int *work_done, int work_to_do)
 891 {
 892         struct net_device *netdev = adapter->netdev;
 893         struct pci_dev *pdev = adapter->pdev;
 894         struct e1000_ring *rx_ring = adapter->rx_ring;
 895         struct e1000_rx_desc *rx_desc, *next_rxd;
 896         struct e1000_buffer *buffer_info, *next_buffer;
 897         u32 length;
 898         unsigned int i;
 899         int cleaned_count = 0;
 900         bool cleaned = false;
 901         unsigned int total_rx_bytes=0, total_rx_packets=0;
 902
 903         i = rx_ring->next_to_clean;
 904         rx_desc = E1000_RX_DESC(*rx_ring, i);
 905         buffer_info = &rx_ring->buffer_info[i];
 906
 907         while (rx_desc->status & E1000_RXD_STAT_DD) {
 908                 struct sk_buff *skb;
 909                 u8 status;
 910
 911                 if (*work_done >= work_to_do)
 912                         break;
 913                 (*work_done)++;
 914
 915                 status = rx_desc->status;
 916                 skb = buffer_info->skb;
 917                 buffer_info->skb = NULL;
 918
 919                 ++i;
 920                 if (i == rx_ring->count)
 921                         i = 0;
 922                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 923                 prefetch(next_rxd);
 924
 925                 next_buffer = &rx_ring->buffer_info[i];
 926
 927                 cleaned = true;
 928                 cleaned_count++;
 929                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 930                                PCI_DMA_FROMDEVICE);
 931                 buffer_info->dma = 0;
 932
 933                 length = le16_to_cpu(rx_desc->length);
 934
 935                 /* errors is only valid for DD + EOP descriptors */
 936                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 937                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 938                                 /* recycle both page and skb */
 939                                 buffer_info->skb = skb;
 940                                 /* an error means any chain goes out the window
 941                                  * too */
 942                                 if (rx_ring->rx_skb_top)
 943                                         dev_kfree_skb(rx_ring->rx_skb_top);
 944                                 rx_ring->rx_skb_top = NULL;
 945                                 goto next_desc;
 946                 }
 947
 948 #define rxtop rx_ring->rx_skb_top
 949                 if (!(status & E1000_RXD_STAT_EOP)) {
 950                         /* this descriptor is only the beginning (or middle) */
 951                         if (!rxtop) {
 952                                 /* this is the beginning of a chain */
 953                                 rxtop = skb;
 954                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 955                                                    0, length);
 956                         } else {
 957                                 /* this is the middle of a chain */
 958                                 skb_fill_page_desc(rxtop,
 959                                     skb_shinfo(rxtop)->nr_frags,
 960                                     buffer_info->page, 0, length);
 961                                 /* re-use the skb, only consumed the page */
 962                                 buffer_info->skb = skb;
 963                         }
 964                         e1000_consume_page(buffer_info, rxtop, length);
 965                         goto next_desc;
 966                 } else {
 967                         if (rxtop) {
 968                                 /* end of the chain */
 969                                 skb_fill_page_desc(rxtop,
 970                                     skb_shinfo(rxtop)->nr_frags,
 971                                     buffer_info->page, 0, length);
 972                                 /* re-use the current skb, we only consumed the
 973                                  * page */
 974                                 buffer_info->skb = skb;
 975                                 skb = rxtop;
 976                                 rxtop = NULL;
 977                                 e1000_consume_page(buffer_info, skb, length);
 978                         } else {
 979                                 /* no chain, got EOP, this buf is the packet
 980                                  * copybreak to save the put_page/alloc_page */
 981                                 if (length <= copybreak &&
 982                                     skb_tailroom(skb) >= length) {
 983                                         u8 *vaddr;
 984                                         vaddr = kmap_atomic(buffer_info->page,
 985                                                            KM_SKB_DATA_SOFTIRQ);
 986                                         memcpy(skb_tail_pointer(skb), vaddr,
 987                                                length);
 988                                         kunmap_atomic(vaddr,
 989                                                       KM_SKB_DATA_SOFTIRQ);
 990                                         /* re-use the page, so don't erase
 991                                          * buffer_info->page */
 992                                         skb_put(skb, length);
 993                                 } else {
 994                                         skb_fill_page_desc(skb, 0,
 995                                                            buffer_info->page, 0,
 996                                                            length);
 997                                         e1000_consume_page(buffer_info, skb,
 998                                                            length);
 999                                 }
1000                         }
1001                 }
1002
1003                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1004                 e1000_rx_checksum(adapter,
1005                                   (u32)(status) |
1006                                   ((u32)(rx_desc->errors) << 24),
1007                                   le16_to_cpu(rx_desc->csum), skb);
1008
1009                 /* probably a little skewed due to removing CRC */
1010                 total_rx_bytes += skb->len;
1011                 total_rx_packets++;
1012
1013                 /* eth type trans needs skb->data to point to something */
1014                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1015                         ndev_err(netdev, "pskb_may_pull failed.\n");
1016                         dev_kfree_skb(skb);
1017                         goto next_desc;
1018                 }
1019
1020                 e1000_receive_skb(adapter, netdev, skb, status,
1021                                   rx_desc->special);
1022
1023 next_desc:
1024                 rx_desc->status = 0;
1025
1026                 /* return some buffers to hardware, one at a time is too slow */
1027                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1028                         adapter->alloc_rx_buf(adapter, cleaned_count);
1029                         cleaned_count = 0;
1030                 }
1031
1032                 /* use prefetched values */
1033                 rx_desc = next_rxd;
1034                 buffer_info = next_buffer;
1035         }
1036         rx_ring->next_to_clean = i;
1037
1038         cleaned_count = e1000_desc_unused(rx_ring);
1039         if (cleaned_count)
1040                 adapter->alloc_rx_buf(adapter, cleaned_count);
1041
1042         adapter->total_rx_bytes += total_rx_bytes;
1043         adapter->total_rx_packets += total_rx_packets;
1044         adapter->net_stats.rx_bytes += total_rx_bytes;
1045         adapter->net_stats.rx_packets += total_rx_packets;
1046         return cleaned;
1047 }
1048
1049 /**
1050  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1051  * @adapter: board private structure
1052  **/
1053 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1054 {
1055         struct e1000_ring *rx_ring = adapter->rx_ring;
1056         struct e1000_buffer *buffer_info;
1057         struct e1000_ps_page *ps_page;
1058         struct pci_dev *pdev = adapter->pdev;
1059         unsigned int i, j;
1060
1061         /* Free all the Rx ring sk_buffs */
1062         for (i = 0; i < rx_ring->count; i++) {
1063                 buffer_info = &rx_ring->buffer_info[i];
1064                 if (buffer_info->dma) {
1065                         if (adapter->clean_rx == e1000_clean_rx_irq)
1066                                 pci_unmap_single(pdev, buffer_info->dma,
1067                                                  adapter->rx_buffer_len,
1068                                                  PCI_DMA_FROMDEVICE);
1069                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1070                                 pci_unmap_page(pdev, buffer_info->dma,
1071                                                PAGE_SIZE,
1072                                                PCI_DMA_FROMDEVICE);
1073                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1074                                 pci_unmap_single(pdev, buffer_info->dma,
1075                                                  adapter->rx_ps_bsize0,
1076                                                  PCI_DMA_FROMDEVICE);
1077                         buffer_info->dma = 0;
1078                 }
1079
1080                 if (buffer_info->page) {
1081                         put_page(buffer_info->page);
1082                         buffer_info->page = NULL;
1083                 }
1084
1085                 if (buffer_info->skb) {
1086                         dev_kfree_skb(buffer_info->skb);
1087                         buffer_info->skb = NULL;
1088                 }
1089
1090                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1091                         ps_page = &buffer_info->ps_pages[j];
1092                         if (!ps_page->page)
1093                                 break;
1094                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1095                                        PCI_DMA_FROMDEVICE);
1096                         ps_page->dma = 0;
1097                         put_page(ps_page->page);
1098                         ps_page->page = NULL;
1099                 }
1100         }
1101
1102         /* there also may be some cached data from a chained receive */
1103         if (rx_ring->rx_skb_top) {
1104                 dev_kfree_skb(rx_ring->rx_skb_top);
1105                 rx_ring->rx_skb_top = NULL;
1106         }
1107
1108         /* Zero out the descriptor ring */
1109         memset(rx_ring->desc, 0, rx_ring->size);
1110
1111         rx_ring->next_to_clean = 0;
1112         rx_ring->next_to_use = 0;
1113
1114         writel(0, adapter->hw.hw_addr + rx_ring->head);
1115         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1116 }
1117
1118 /**
1119  * e1000_intr_msi - Interrupt Handler
1120  * @irq: interrupt number
1121  * @data: pointer to a network interface device structure
1122  **/
1123 static irqreturn_t e1000_intr_msi(int irq, void *data)
1124 {
1125         struct net_device *netdev = data;
1126         struct e1000_adapter *adapter = netdev_priv(netdev);
1127         struct e1000_hw *hw = &adapter->hw;
1128         u32 icr = er32(ICR);
1129
1130         /*
1131          * read ICR disables interrupts using IAM
1132          */
1133
1134         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1135                 hw->mac.get_link_status = 1;
1136                 /*
1137                  * ICH8 workaround-- Call gig speed drop workaround on cable
1138                  * disconnect (LSC) before accessing any PHY registers
1139                  */
1140                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1141                     (!(er32(STATUS) & E1000_STATUS_LU)))
1142                         e1000e_gig_downshift_workaround_ich8lan(hw);
1143
1144                 /*
1145                  * 80003ES2LAN workaround-- For packet buffer work-around on
1146                  * link down event; disable receives here in the ISR and reset
1147                  * adapter in watchdog
1148                  */
1149                 if (netif_carrier_ok(netdev) &&
1150                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1151                         /* disable receives */
1152                         u32 rctl = er32(RCTL);
1153                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1154                         adapter->flags |= FLAG_RX_RESTART_NOW;
1155                 }
1156                 /* guard against interrupt when we're going down */
1157                 if (!test_bit(__E1000_DOWN, &adapter->state))
1158                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1159         }
1160
1161         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1162                 adapter->total_tx_bytes = 0;
1163                 adapter->total_tx_packets = 0;
1164                 adapter->total_rx_bytes = 0;
1165                 adapter->total_rx_packets = 0;
1166                 __netif_rx_schedule(netdev, &adapter->napi);
1167         }
1168
1169         return IRQ_HANDLED;
1170 }
1171
1172 /**
1173  * e1000_intr - Interrupt Handler
1174  * @irq: interrupt number
1175  * @data: pointer to a network interface device structure
1176  **/
1177 static irqreturn_t e1000_intr(int irq, void *data)
1178 {
1179         struct net_device *netdev = data;
1180         struct e1000_adapter *adapter = netdev_priv(netdev);
1181         struct e1000_hw *hw = &adapter->hw;
1182
1183         u32 rctl, icr = er32(ICR);
1184         if (!icr)
1185                 return IRQ_NONE;  /* Not our interrupt */
1186
1187         /*
1188          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1189          * not set, then the adapter didn't send an interrupt
1190          */
1191         if (!(icr & E1000_ICR_INT_ASSERTED))
1192                 return IRQ_NONE;
1193
1194         /*
1195          * Interrupt Auto-Mask...upon reading ICR,
1196          * interrupts are masked.  No need for the
1197          * IMC write
1198          */
1199
1200         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1201                 hw->mac.get_link_status = 1;
1202                 /*
1203                  * ICH8 workaround-- Call gig speed drop workaround on cable
1204                  * disconnect (LSC) before accessing any PHY registers
1205                  */
1206                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1207                     (!(er32(STATUS) & E1000_STATUS_LU)))
1208                         e1000e_gig_downshift_workaround_ich8lan(hw);
1209
1210                 /*
1211                  * 80003ES2LAN workaround--
1212                  * For packet buffer work-around on link down event;
1213                  * disable receives here in the ISR and
1214                  * reset adapter in watchdog
1215                  */
1216                 if (netif_carrier_ok(netdev) &&
1217                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1218                         /* disable receives */
1219                         rctl = er32(RCTL);
1220                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1221                         adapter->flags |= FLAG_RX_RESTART_NOW;
1222                 }
1223                 /* guard against interrupt when we're going down */
1224                 if (!test_bit(__E1000_DOWN, &adapter->state))
1225                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1226         }
1227
1228         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1229                 adapter->total_tx_bytes = 0;
1230                 adapter->total_tx_packets = 0;
1231                 adapter->total_rx_bytes = 0;
1232                 adapter->total_rx_packets = 0;
1233                 __netif_rx_schedule(netdev, &adapter->napi);
1234         }
1235
1236         return IRQ_HANDLED;
1237 }
1238
1239 static int e1000_request_irq(struct e1000_adapter *adapter)
1240 {
1241         struct net_device *netdev = adapter->netdev;
1242         irq_handler_t handler = e1000_intr;
1243         int irq_flags = IRQF_SHARED;
1244         int err;
1245
1246         if (!pci_enable_msi(adapter->pdev)) {
1247                 adapter->flags |= FLAG_MSI_ENABLED;
1248                 handler = e1000_intr_msi;
1249                 irq_flags = 0;
1250         }
1251
1252         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
1253                           netdev);
1254         if (err) {
1255                 ndev_err(netdev,
1256                        "Unable to allocate %s interrupt (return: %d)\n",
1257                         adapter->flags & FLAG_MSI_ENABLED ? "MSI":"INTx",
1258                         err);
1259                 if (adapter->flags & FLAG_MSI_ENABLED)
1260                         pci_disable_msi(adapter->pdev);
1261         }
1262
1263         return err;
1264 }
1265
1266 static void e1000_free_irq(struct e1000_adapter *adapter)
1267 {
1268         struct net_device *netdev = adapter->netdev;
1269
1270         free_irq(adapter->pdev->irq, netdev);
1271         if (adapter->flags & FLAG_MSI_ENABLED) {
1272                 pci_disable_msi(adapter->pdev);
1273                 adapter->flags &= ~FLAG_MSI_ENABLED;
1274         }
1275 }
1276
1277 /**
1278  * e1000_irq_disable - Mask off interrupt generation on the NIC
1279  **/
1280 static void e1000_irq_disable(struct e1000_adapter *adapter)
1281 {
1282         struct e1000_hw *hw = &adapter->hw;
1283
1284         ew32(IMC, ~0);
1285         e1e_flush();
1286         synchronize_irq(adapter->pdev->irq);
1287 }
1288
1289 /**
1290  * e1000_irq_enable - Enable default interrupt generation settings
1291  **/
1292 static void e1000_irq_enable(struct e1000_adapter *adapter)
1293 {
1294         struct e1000_hw *hw = &adapter->hw;
1295
1296         ew32(IMS, IMS_ENABLE_MASK);
1297         e1e_flush();
1298 }
1299
1300 /**
1301  * e1000_get_hw_control - get control of the h/w from f/w
1302  * @adapter: address of board private structure
1303  *
1304  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1305  * For ASF and Pass Through versions of f/w this means that
1306  * the driver is loaded. For AMT version (only with 82573)
1307  * of the f/w this means that the network i/f is open.
1308  **/
1309 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1310 {
1311         struct e1000_hw *hw = &adapter->hw;
1312         u32 ctrl_ext;
1313         u32 swsm;
1314
1315         /* Let firmware know the driver has taken over */
1316         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1317                 swsm = er32(SWSM);
1318                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1319         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1320                 ctrl_ext = er32(CTRL_EXT);
1321                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1322         }
1323 }
1324
1325 /**
1326  * e1000_release_hw_control - release control of the h/w to f/w
1327  * @adapter: address of board private structure
1328  *
1329  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1330  * For ASF and Pass Through versions of f/w this means that the
1331  * driver is no longer loaded. For AMT version (only with 82573) i
1332  * of the f/w this means that the network i/f is closed.
1333  *
1334  **/
1335 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1336 {
1337         struct e1000_hw *hw = &adapter->hw;
1338         u32 ctrl_ext;
1339         u32 swsm;
1340
1341         /* Let firmware taken over control of h/w */
1342         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1343                 swsm = er32(SWSM);
1344                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1345         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1346                 ctrl_ext = er32(CTRL_EXT);
1347                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1348         }
1349 }
1350
1351 /**
1352  * @e1000_alloc_ring - allocate memory for a ring structure
1353  **/
1354 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1355                                 struct e1000_ring *ring)
1356 {
1357         struct pci_dev *pdev = adapter->pdev;
1358
1359         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1360                                         GFP_KERNEL);
1361         if (!ring->desc)
1362                 return -ENOMEM;
1363
1364         return 0;
1365 }
1366
1367 /**
1368  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1369  * @adapter: board private structure
1370  *
1371  * Return 0 on success, negative on failure
1372  **/
1373 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1374 {
1375         struct e1000_ring *tx_ring = adapter->tx_ring;
1376         int err = -ENOMEM, size;
1377
1378         size = sizeof(struct e1000_buffer) * tx_ring->count;
1379         tx_ring->buffer_info = vmalloc(size);
1380         if (!tx_ring->buffer_info)
1381                 goto err;
1382         memset(tx_ring->buffer_info, 0, size);
1383
1384         /* round up to nearest 4K */
1385         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1386         tx_ring->size = ALIGN(tx_ring->size, 4096);
1387
1388         err = e1000_alloc_ring_dma(adapter, tx_ring);
1389         if (err)
1390                 goto err;
1391
1392         tx_ring->next_to_use = 0;
1393         tx_ring->next_to_clean = 0;
1394         spin_lock_init(&adapter->tx_queue_lock);
1395
1396         return 0;
1397 err:
1398         vfree(tx_ring->buffer_info);
1399         ndev_err(adapter->netdev,
1400         "Unable to allocate memory for the transmit descriptor ring\n");
1401         return err;
1402 }
1403
1404 /**
1405  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1406  * @adapter: board private structure
1407  *
1408  * Returns 0 on success, negative on failure
1409  **/
1410 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1411 {
1412         struct e1000_ring *rx_ring = adapter->rx_ring;
1413         struct e1000_buffer *buffer_info;
1414         int i, size, desc_len, err = -ENOMEM;
1415
1416         size = sizeof(struct e1000_buffer) * rx_ring->count;
1417         rx_ring->buffer_info = vmalloc(size);
1418         if (!rx_ring->buffer_info)
1419                 goto err;
1420         memset(rx_ring->buffer_info, 0, size);
1421
1422         for (i = 0; i < rx_ring->count; i++) {
1423                 buffer_info = &rx_ring->buffer_info[i];
1424                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1425                                                 sizeof(struct e1000_ps_page),
1426                                                 GFP_KERNEL);
1427                 if (!buffer_info->ps_pages)
1428                         goto err_pages;
1429         }
1430
1431         desc_len = sizeof(union e1000_rx_desc_packet_split);
1432
1433         /* Round up to nearest 4K */
1434         rx_ring->size = rx_ring->count * desc_len;
1435         rx_ring->size = ALIGN(rx_ring->size, 4096);
1436
1437         err = e1000_alloc_ring_dma(adapter, rx_ring);
1438         if (err)
1439                 goto err_pages;
1440
1441         rx_ring->next_to_clean = 0;
1442         rx_ring->next_to_use = 0;
1443         rx_ring->rx_skb_top = NULL;
1444
1445         return 0;
1446
1447 err_pages:
1448         for (i = 0; i < rx_ring->count; i++) {
1449                 buffer_info = &rx_ring->buffer_info[i];
1450                 kfree(buffer_info->ps_pages);
1451         }
1452 err:
1453         vfree(rx_ring->buffer_info);
1454         ndev_err(adapter->netdev,
1455         "Unable to allocate memory for the transmit descriptor ring\n");
1456         return err;
1457 }
1458
1459 /**
1460  * e1000_clean_tx_ring - Free Tx Buffers
1461  * @adapter: board private structure
1462  **/
1463 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1464 {
1465         struct e1000_ring *tx_ring = adapter->tx_ring;
1466         struct e1000_buffer *buffer_info;
1467         unsigned long size;
1468         unsigned int i;
1469
1470         for (i = 0; i < tx_ring->count; i++) {
1471                 buffer_info = &tx_ring->buffer_info[i];
1472                 e1000_put_txbuf(adapter, buffer_info);
1473         }
1474
1475         size = sizeof(struct e1000_buffer) * tx_ring->count;
1476         memset(tx_ring->buffer_info, 0, size);
1477
1478         memset(tx_ring->desc, 0, tx_ring->size);
1479
1480         tx_ring->next_to_use = 0;
1481         tx_ring->next_to_clean = 0;
1482
1483         writel(0, adapter->hw.hw_addr + tx_ring->head);
1484         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1485 }
1486
1487 /**
1488  * e1000e_free_tx_resources - Free Tx Resources per Queue
1489  * @adapter: board private structure
1490  *
1491  * Free all transmit software resources
1492  **/
1493 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1494 {
1495         struct pci_dev *pdev = adapter->pdev;
1496         struct e1000_ring *tx_ring = adapter->tx_ring;
1497
1498         e1000_clean_tx_ring(adapter);
1499
1500         vfree(tx_ring->buffer_info);
1501         tx_ring->buffer_info = NULL;
1502
1503         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1504                           tx_ring->dma);
1505         tx_ring->desc = NULL;
1506 }
1507
1508 /**
1509  * e1000e_free_rx_resources - Free Rx Resources
1510  * @adapter: board private structure
1511  *
1512  * Free all receive software resources
1513  **/
1514
1515 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1516 {
1517         struct pci_dev *pdev = adapter->pdev;
1518         struct e1000_ring *rx_ring = adapter->rx_ring;
1519         int i;
1520
1521         e1000_clean_rx_ring(adapter);
1522
1523         for (i = 0; i < rx_ring->count; i++) {
1524                 kfree(rx_ring->buffer_info[i].ps_pages);
1525         }
1526
1527         vfree(rx_ring->buffer_info);
1528         rx_ring->buffer_info = NULL;
1529
1530         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1531                           rx_ring->dma);
1532         rx_ring->desc = NULL;
1533 }
1534
1535 /**
1536  * e1000_update_itr - update the dynamic ITR value based on statistics
1537  * @adapter: pointer to adapter
1538  * @itr_setting: current adapter->itr
1539  * @packets: the number of packets during this measurement interval
1540  * @bytes: the number of bytes during this measurement interval
1541  *
1542  *      Stores a new ITR value based on packets and byte
1543  *      counts during the last interrupt.  The advantage of per interrupt
1544  *      computation is faster updates and more accurate ITR for the current
1545  *      traffic pattern.  Constants in this function were computed
1546  *      based on theoretical maximum wire speed and thresholds were set based
1547  *      on testing data as well as attempting to minimize response time
1548  *      while increasing bulk throughput.
1549  *      this functionality is controlled by the InterruptThrottleRate module
1550  *      parameter (see e1000_param.c)
1551  **/
1552 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1553                                      u16 itr_setting, int packets,
1554                                      int bytes)
1555 {
1556         unsigned int retval = itr_setting;
1557
1558         if (packets == 0)
1559                 goto update_itr_done;
1560
1561         switch (itr_setting) {
1562         case lowest_latency:
1563                 /* handle TSO and jumbo frames */
1564                 if (bytes/packets > 8000)
1565                         retval = bulk_latency;
1566                 else if ((packets < 5) && (bytes > 512)) {
1567                         retval = low_latency;
1568                 }
1569                 break;
1570         case low_latency:  /* 50 usec aka 20000 ints/s */
1571                 if (bytes > 10000) {
1572                         /* this if handles the TSO accounting */
1573                         if (bytes/packets > 8000) {
1574                                 retval = bulk_latency;
1575                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1576                                 retval = bulk_latency;
1577                         } else if ((packets > 35)) {
1578                                 retval = lowest_latency;
1579                         }
1580                 } else if (bytes/packets > 2000) {
1581                         retval = bulk_latency;
1582                 } else if (packets <= 2 && bytes < 512) {
1583                         retval = lowest_latency;
1584                 }
1585                 break;
1586         case bulk_latency: /* 250 usec aka 4000 ints/s */
1587                 if (bytes > 25000) {
1588                         if (packets > 35) {
1589                                 retval = low_latency;
1590                         }
1591                 } else if (bytes < 6000) {
1592                         retval = low_latency;
1593                 }
1594                 break;
1595         }
1596
1597 update_itr_done:
1598         return retval;
1599 }
1600
1601 static void e1000_set_itr(struct e1000_adapter *adapter)
1602 {
1603         struct e1000_hw *hw = &adapter->hw;
1604         u16 current_itr;
1605         u32 new_itr = adapter->itr;
1606
1607         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1608         if (adapter->link_speed != SPEED_1000) {
1609                 current_itr = 0;
1610                 new_itr = 4000;
1611                 goto set_itr_now;
1612         }
1613
1614         adapter->tx_itr = e1000_update_itr(adapter,
1615                                     adapter->tx_itr,
1616                                     adapter->total_tx_packets,
1617                                     adapter->total_tx_bytes);
1618         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1619         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1620                 adapter->tx_itr = low_latency;
1621
1622         adapter->rx_itr = e1000_update_itr(adapter,
1623                                     adapter->rx_itr,
1624                                     adapter->total_rx_packets,
1625                                     adapter->total_rx_bytes);
1626         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1627         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1628                 adapter->rx_itr = low_latency;
1629
1630         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1631
1632         switch (current_itr) {
1633         /* counts and packets in update_itr are dependent on these numbers */
1634         case lowest_latency:
1635                 new_itr = 70000;
1636                 break;
1637         case low_latency:
1638                 new_itr = 20000; /* aka hwitr = ~200 */
1639                 break;
1640         case bulk_latency:
1641                 new_itr = 4000;
1642                 break;
1643         default:
1644                 break;
1645         }
1646
1647 set_itr_now:
1648         if (new_itr != adapter->itr) {
1649                 /*
1650                  * this attempts to bias the interrupt rate towards Bulk
1651                  * by adding intermediate steps when interrupt rate is
1652                  * increasing
1653                  */
1654                 new_itr = new_itr > adapter->itr ?
1655                              min(adapter->itr + (new_itr >> 2), new_itr) :
1656                              new_itr;
1657                 adapter->itr = new_itr;
1658                 ew32(ITR, 1000000000 / (new_itr * 256));
1659         }
1660 }
1661
1662 /**
1663  * e1000_clean - NAPI Rx polling callback
1664  * @napi: struct associated with this polling callback
1665  * @budget: amount of packets driver is allowed to process this poll
1666  **/
1667 static int e1000_clean(struct napi_struct *napi, int budget)
1668 {
1669         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1670         struct net_device *poll_dev = adapter->netdev;
1671         int tx_cleaned = 0, work_done = 0;
1672
1673         /* Must NOT use netdev_priv macro here. */
1674         adapter = poll_dev->priv;
1675
1676         /*
1677          * e1000_clean is called per-cpu.  This lock protects
1678          * tx_ring from being cleaned by multiple cpus
1679          * simultaneously.  A failure obtaining the lock means
1680          * tx_ring is currently being cleaned anyway.
1681          */
1682         if (spin_trylock(&adapter->tx_queue_lock)) {
1683                 tx_cleaned = e1000_clean_tx_irq(adapter);
1684                 spin_unlock(&adapter->tx_queue_lock);
1685         }
1686
1687         adapter->clean_rx(adapter, &work_done, budget);
1688
1689         if (tx_cleaned)
1690                 work_done = budget;
1691
1692         /* If budget not fully consumed, exit the polling mode */
1693         if (work_done < budget) {
1694                 if (adapter->itr_setting & 3)
1695                         e1000_set_itr(adapter);
1696                 netif_rx_complete(poll_dev, napi);
1697                 e1000_irq_enable(adapter);
1698         }
1699
1700         return work_done;
1701 }
1702
1703 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1704 {
1705         struct e1000_adapter *adapter = netdev_priv(netdev);
1706         struct e1000_hw *hw = &adapter->hw;
1707         u32 vfta, index;
1708
1709         /* don't update vlan cookie if already programmed */
1710         if ((adapter->hw.mng_cookie.status &
1711              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1712             (vid == adapter->mng_vlan_id))
1713                 return;
1714         /* add VID to filter table */
1715         index = (vid >> 5) & 0x7F;
1716         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1717         vfta |= (1 << (vid & 0x1F));
1718         e1000e_write_vfta(hw, index, vfta);
1719 }
1720
1721 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1722 {
1723         struct e1000_adapter *adapter = netdev_priv(netdev);
1724         struct e1000_hw *hw = &adapter->hw;
1725         u32 vfta, index;
1726
1727         if (!test_bit(__E1000_DOWN, &adapter->state))
1728                 e1000_irq_disable(adapter);
1729         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1730
1731         if (!test_bit(__E1000_DOWN, &adapter->state))
1732                 e1000_irq_enable(adapter);
1733
1734         if ((adapter->hw.mng_cookie.status &
1735              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1736             (vid == adapter->mng_vlan_id)) {
1737                 /* release control to f/w */
1738                 e1000_release_hw_control(adapter);
1739                 return;
1740         }
1741
1742         /* remove VID from filter table */
1743         index = (vid >> 5) & 0x7F;
1744         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1745         vfta &= ~(1 << (vid & 0x1F));
1746         e1000e_write_vfta(hw, index, vfta);
1747 }
1748
1749 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1750 {
1751         struct net_device *netdev = adapter->netdev;
1752         u16 vid = adapter->hw.mng_cookie.vlan_id;
1753         u16 old_vid = adapter->mng_vlan_id;
1754
1755         if (!adapter->vlgrp)
1756                 return;
1757
1758         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1759                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1760                 if (adapter->hw.mng_cookie.status &
1761                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1762                         e1000_vlan_rx_add_vid(netdev, vid);
1763                         adapter->mng_vlan_id = vid;
1764                 }
1765
1766                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1767                                 (vid != old_vid) &&
1768                     !vlan_group_get_device(adapter->vlgrp, old_vid))
1769                         e1000_vlan_rx_kill_vid(netdev, old_vid);
1770         } else {
1771                 adapter->mng_vlan_id = vid;
1772         }
1773 }
1774
1775
1776 static void e1000_vlan_rx_register(struct net_device *netdev,
1777                                    struct vlan_group *grp)
1778 {
1779         struct e1000_adapter *adapter = netdev_priv(netdev);
1780         struct e1000_hw *hw = &adapter->hw;
1781         u32 ctrl, rctl;
1782
1783         if (!test_bit(__E1000_DOWN, &adapter->state))
1784                 e1000_irq_disable(adapter);
1785         adapter->vlgrp = grp;
1786
1787         if (grp) {
1788                 /* enable VLAN tag insert/strip */
1789                 ctrl = er32(CTRL);
1790                 ctrl |= E1000_CTRL_VME;
1791                 ew32(CTRL, ctrl);
1792
1793                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1794                         /* enable VLAN receive filtering */
1795                         rctl = er32(RCTL);
1796                         rctl |= E1000_RCTL_VFE;
1797                         rctl &= ~E1000_RCTL_CFIEN;
1798                         ew32(RCTL, rctl);
1799                         e1000_update_mng_vlan(adapter);
1800                 }
1801         } else {
1802                 /* disable VLAN tag insert/strip */
1803                 ctrl = er32(CTRL);
1804                 ctrl &= ~E1000_CTRL_VME;
1805                 ew32(CTRL, ctrl);
1806
1807                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1808                         /* disable VLAN filtering */
1809                         rctl = er32(RCTL);
1810                         rctl &= ~E1000_RCTL_VFE;
1811                         ew32(RCTL, rctl);
1812                         if (adapter->mng_vlan_id !=
1813                             (u16)E1000_MNG_VLAN_NONE) {
1814                                 e1000_vlan_rx_kill_vid(netdev,
1815                                                        adapter->mng_vlan_id);
1816                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1817                         }
1818                 }
1819         }
1820
1821         if (!test_bit(__E1000_DOWN, &adapter->state))
1822                 e1000_irq_enable(adapter);
1823 }
1824
1825 static void e1000_restore_vlan(struct e1000_adapter *adapter)
1826 {
1827         u16 vid;
1828
1829         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1830
1831         if (!adapter->vlgrp)
1832                 return;
1833
1834         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1835                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1836                         continue;
1837                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1838         }
1839 }
1840
1841 static void e1000_init_manageability(struct e1000_adapter *adapter)
1842 {
1843         struct e1000_hw *hw = &adapter->hw;
1844         u32 manc, manc2h;
1845
1846         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1847                 return;
1848
1849         manc = er32(MANC);
1850
1851         /*
1852          * enable receiving management packets to the host. this will probably
1853          * generate destination unreachable messages from the host OS, but
1854          * the packets will be handled on SMBUS
1855          */
1856         manc |= E1000_MANC_EN_MNG2HOST;
1857         manc2h = er32(MANC2H);
1858 #define E1000_MNG2HOST_PORT_623 (1 << 5)
1859 #define E1000_MNG2HOST_PORT_664 (1 << 6)
1860         manc2h |= E1000_MNG2HOST_PORT_623;
1861         manc2h |= E1000_MNG2HOST_PORT_664;
1862         ew32(MANC2H, manc2h);
1863         ew32(MANC, manc);
1864 }
1865
1866 /**
1867  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1868  * @adapter: board private structure
1869  *
1870  * Configure the Tx unit of the MAC after a reset.
1871  **/
1872 static void e1000_configure_tx(struct e1000_adapter *adapter)
1873 {
1874         struct e1000_hw *hw = &adapter->hw;
1875         struct e1000_ring *tx_ring = adapter->tx_ring;
1876         u64 tdba;
1877         u32 tdlen, tctl, tipg, tarc;
1878         u32 ipgr1, ipgr2;
1879
1880         /* Setup the HW Tx Head and Tail descriptor pointers */
1881         tdba = tx_ring->dma;
1882         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1883         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1884         ew32(TDBAH, (tdba >> 32));
1885         ew32(TDLEN, tdlen);
1886         ew32(TDH, 0);
1887         ew32(TDT, 0);
1888         tx_ring->head = E1000_TDH;
1889         tx_ring->tail = E1000_TDT;
1890
1891         /* Set the default values for the Tx Inter Packet Gap timer */
1892         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
1893         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
1894         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
1895
1896         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1897                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
1898
1899         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1900         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1901         ew32(TIPG, tipg);
1902
1903         /* Set the Tx Interrupt Delay register */
1904         ew32(TIDV, adapter->tx_int_delay);
1905         /* Tx irq moderation */
1906         ew32(TADV, adapter->tx_abs_int_delay);
1907
1908         /* Program the Transmit Control Register */
1909         tctl = er32(TCTL);
1910         tctl &= ~E1000_TCTL_CT;
1911         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1912                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1913
1914         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1915                 tarc = er32(TARC(0));
1916                 /*
1917                  * set the speed mode bit, we'll clear it if we're not at
1918                  * gigabit link later
1919                  */
1920 #define SPEED_MODE_BIT (1 << 21)
1921                 tarc |= SPEED_MODE_BIT;
1922                 ew32(TARC(0), tarc);
1923         }
1924
1925         /* errata: program both queues to unweighted RR */
1926         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1927                 tarc = er32(TARC(0));
1928                 tarc |= 1;
1929                 ew32(TARC(0), tarc);
1930                 tarc = er32(TARC(1));
1931                 tarc |= 1;
1932                 ew32(TARC(1), tarc);
1933         }
1934
1935         e1000e_config_collision_dist(hw);
1936
1937         /* Setup Transmit Descriptor Settings for eop descriptor */
1938         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1939
1940         /* only set IDE if we are delaying interrupts using the timers */
1941         if (adapter->tx_int_delay)
1942                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1943
1944         /* enable Report Status bit */
1945         adapter->txd_cmd |= E1000_TXD_CMD_RS;
1946
1947         ew32(TCTL, tctl);
1948
1949         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1950 }
1951
1952 /**
1953  * e1000_setup_rctl - configure the receive control registers
1954  * @adapter: Board private structure
1955  **/
1956 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1957                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1958 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1959 {
1960         struct e1000_hw *hw = &adapter->hw;
1961         u32 rctl, rfctl;
1962         u32 psrctl = 0;
1963         u32 pages = 0;
1964
1965         /* Program MC offset vector base */
1966         rctl = er32(RCTL);
1967         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1968         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1969                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1970                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1971
1972         /* Do not Store bad packets */
1973         rctl &= ~E1000_RCTL_SBP;
1974
1975         /* Enable Long Packet receive */
1976         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1977                 rctl &= ~E1000_RCTL_LPE;
1978         else
1979                 rctl |= E1000_RCTL_LPE;
1980
1981         /* Enable hardware CRC frame stripping */
1982         rctl |= E1000_RCTL_SECRC;
1983
1984         /* Setup buffer sizes */
1985         rctl &= ~E1000_RCTL_SZ_4096;
1986         rctl |= E1000_RCTL_BSEX;
1987         switch (adapter->rx_buffer_len) {
1988         case 256:
1989                 rctl |= E1000_RCTL_SZ_256;
1990                 rctl &= ~E1000_RCTL_BSEX;
1991                 break;
1992         case 512:
1993                 rctl |= E1000_RCTL_SZ_512;
1994                 rctl &= ~E1000_RCTL_BSEX;
1995                 break;
1996         case 1024:
1997                 rctl |= E1000_RCTL_SZ_1024;
1998                 rctl &= ~E1000_RCTL_BSEX;
1999                 break;
2000         case 2048:
2001         default:
2002                 rctl |= E1000_RCTL_SZ_2048;
2003                 rctl &= ~E1000_RCTL_BSEX;
2004                 break;
2005         case 4096:
2006                 rctl |= E1000_RCTL_SZ_4096;
2007                 break;
2008         case 8192:
2009                 rctl |= E1000_RCTL_SZ_8192;
2010                 break;
2011         case 16384:
2012                 rctl |= E1000_RCTL_SZ_16384;
2013                 break;
2014         }
2015
2016         /*
2017          * 82571 and greater support packet-split where the protocol
2018          * header is placed in skb->data and the packet data is
2019          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2020          * In the case of a non-split, skb->data is linearly filled,
2021          * followed by the page buffers.  Therefore, skb->data is
2022          * sized to hold the largest protocol header.
2023          *
2024          * allocations using alloc_page take too long for regular MTU
2025          * so only enable packet split for jumbo frames
2026          *
2027          * Using pages when the page size is greater than 16k wastes
2028          * a lot of memory, since we allocate 3 pages at all times
2029          * per packet.
2030          */
2031         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2032         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2033             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2034                 adapter->rx_ps_pages = pages;
2035         else
2036                 adapter->rx_ps_pages = 0;
2037
2038         if (adapter->rx_ps_pages) {
2039                 /* Configure extra packet-split registers */
2040                 rfctl = er32(RFCTL);
2041                 rfctl |= E1000_RFCTL_EXTEN;
2042                 /*
2043                  * disable packet split support for IPv6 extension headers,
2044                  * because some malformed IPv6 headers can hang the Rx
2045                  */
2046                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2047                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2048
2049                 ew32(RFCTL, rfctl);
2050
2051                 /* Enable Packet split descriptors */
2052                 rctl |= E1000_RCTL_DTYP_PS;
2053
2054                 psrctl |= adapter->rx_ps_bsize0 >>
2055                         E1000_PSRCTL_BSIZE0_SHIFT;
2056
2057                 switch (adapter->rx_ps_pages) {
2058                 case 3:
2059                         psrctl |= PAGE_SIZE <<
2060                                 E1000_PSRCTL_BSIZE3_SHIFT;
2061                 case 2:
2062                         psrctl |= PAGE_SIZE <<
2063                                 E1000_PSRCTL_BSIZE2_SHIFT;
2064                 case 1:
2065                         psrctl |= PAGE_SIZE >>
2066                                 E1000_PSRCTL_BSIZE1_SHIFT;
2067                         break;
2068                 }
2069
2070                 ew32(PSRCTL, psrctl);
2071         }
2072
2073         ew32(RCTL, rctl);
2074         /* just started the receive unit, no need to restart */
2075         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2076 }
2077
2078 /**
2079  * e1000_configure_rx - Configure Receive Unit after Reset
2080  * @adapter: board private structure
2081  *
2082  * Configure the Rx unit of the MAC after a reset.
2083  **/
2084 static void e1000_configure_rx(struct e1000_adapter *adapter)
2085 {
2086         struct e1000_hw *hw = &adapter->hw;
2087         struct e1000_ring *rx_ring = adapter->rx_ring;
2088         u64 rdba;
2089         u32 rdlen, rctl, rxcsum, ctrl_ext;
2090
2091         if (adapter->rx_ps_pages) {
2092                 /* this is a 32 byte descriptor */
2093                 rdlen = rx_ring->count *
2094                         sizeof(union e1000_rx_desc_packet_split);
2095                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2096                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2097         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2098                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2099                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2100                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2101         } else {
2102                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2103                 adapter->clean_rx = e1000_clean_rx_irq;
2104                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2105         }
2106
2107         /* disable receives while setting up the descriptors */
2108         rctl = er32(RCTL);
2109         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2110         e1e_flush();
2111         msleep(10);
2112
2113         /* set the Receive Delay Timer Register */
2114         ew32(RDTR, adapter->rx_int_delay);
2115
2116         /* irq moderation */
2117         ew32(RADV, adapter->rx_abs_int_delay);
2118         if (adapter->itr_setting != 0)
2119                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2120
2121         ctrl_ext = er32(CTRL_EXT);
2122         /* Reset delay timers after every interrupt */
2123         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2124         /* Auto-Mask interrupts upon ICR access */
2125         ctrl_ext |= E1000_CTRL_EXT_IAME;
2126         ew32(IAM, 0xffffffff);
2127         ew32(CTRL_EXT, ctrl_ext);
2128         e1e_flush();
2129
2130         /*
2131          * Setup the HW Rx Head and Tail Descriptor Pointers and
2132          * the Base and Length of the Rx Descriptor Ring
2133          */
2134         rdba = rx_ring->dma;
2135         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2136         ew32(RDBAH, (rdba >> 32));
2137         ew32(RDLEN, rdlen);
2138         ew32(RDH, 0);
2139         ew32(RDT, 0);
2140         rx_ring->head = E1000_RDH;
2141         rx_ring->tail = E1000_RDT;
2142
2143         /* Enable Receive Checksum Offload for TCP and UDP */
2144         rxcsum = er32(RXCSUM);
2145         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2146                 rxcsum |= E1000_RXCSUM_TUOFL;
2147
2148                 /*
2149                  * IPv4 payload checksum for UDP fragments must be
2150                  * used in conjunction with packet-split.
2151                  */
2152                 if (adapter->rx_ps_pages)
2153                         rxcsum |= E1000_RXCSUM_IPPCSE;
2154         } else {
2155                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2156                 /* no need to clear IPPCSE as it defaults to 0 */
2157         }
2158         ew32(RXCSUM, rxcsum);
2159
2160         /*
2161          * Enable early receives on supported devices, only takes effect when
2162          * packet size is equal or larger than the specified value (in 8 byte
2163          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2164          */
2165         if ((adapter->flags & FLAG_HAS_ERT) &&
2166             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2167                 u32 rxdctl = er32(RXDCTL(0));
2168                 ew32(RXDCTL(0), rxdctl | 0x3);
2169                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2170                 /*
2171                  * With jumbo frames and early-receive enabled, excessive
2172                  * C4->C2 latencies result in dropped transactions.
2173                  */
2174                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2175                                           e1000e_driver_name, 55);
2176         } else {
2177                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2178                                           e1000e_driver_name,
2179                                           PM_QOS_DEFAULT_VALUE);
2180         }
2181
2182         /* Enable Receives */
2183         ew32(RCTL, rctl);
2184 }
2185
2186 /**
2187  *  e1000_update_mc_addr_list - Update Multicast addresses
2188  *  @hw: pointer to the HW structure
2189  *  @mc_addr_list: array of multicast addresses to program
2190  *  @mc_addr_count: number of multicast addresses to program
2191  *  @rar_used_count: the first RAR register free to program
2192  *  @rar_count: total number of supported Receive Address Registers
2193  *
2194  *  Updates the Receive Address Registers and Multicast Table Array.
2195  *  The caller must have a packed mc_addr_list of multicast addresses.
2196  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2197  *  unless there are workarounds that change this.  Currently no func pointer
2198  *  exists and all implementations are handled in the generic version of this
2199  *  function.
2200  **/
2201 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2202                                       u32 mc_addr_count, u32 rar_used_count,
2203                                       u32 rar_count)
2204 {
2205         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2206                                         rar_used_count, rar_count);
2207 }
2208
2209 /**
2210  * e1000_set_multi - Multicast and Promiscuous mode set
2211  * @netdev: network interface device structure
2212  *
2213  * The set_multi entry point is called whenever the multicast address
2214  * list or the network interface flags are updated.  This routine is
2215  * responsible for configuring the hardware for proper multicast,
2216  * promiscuous mode, and all-multi behavior.
2217  **/
2218 static void e1000_set_multi(struct net_device *netdev)
2219 {
2220         struct e1000_adapter *adapter = netdev_priv(netdev);
2221         struct e1000_hw *hw = &adapter->hw;
2222         struct e1000_mac_info *mac = &hw->mac;
2223         struct dev_mc_list *mc_ptr;
2224         u8  *mta_list;
2225         u32 rctl;
2226         int i;
2227
2228         /* Check for Promiscuous and All Multicast modes */
2229
2230         rctl = er32(RCTL);
2231
2232         if (netdev->flags & IFF_PROMISC) {
2233                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2234         } else if (netdev->flags & IFF_ALLMULTI) {
2235                 rctl |= E1000_RCTL_MPE;
2236                 rctl &= ~E1000_RCTL_UPE;
2237         } else {
2238                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2239         }
2240
2241         ew32(RCTL, rctl);
2242
2243         if (netdev->mc_count) {
2244                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2245                 if (!mta_list)
2246                         return;
2247
2248                 /* prepare a packed array of only addresses. */
2249                 mc_ptr = netdev->mc_list;
2250
2251                 for (i = 0; i < netdev->mc_count; i++) {
2252                         if (!mc_ptr)
2253                                 break;
2254                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2255                                ETH_ALEN);
2256                         mc_ptr = mc_ptr->next;
2257                 }
2258
2259                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2260                                           mac->rar_entry_count);
2261                 kfree(mta_list);
2262         } else {
2263                 /*
2264                  * if we're called from probe, we might not have
2265                  * anything to do here, so clear out the list
2266                  */
2267                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2268         }
2269 }
2270
2271 /**
2272  * e1000_configure - configure the hardware for Rx and Tx
2273  * @adapter: private board structure
2274  **/
2275 static void e1000_configure(struct e1000_adapter *adapter)
2276 {
2277         e1000_set_multi(adapter->netdev);
2278
2279         e1000_restore_vlan(adapter);
2280         e1000_init_manageability(adapter);
2281
2282         e1000_configure_tx(adapter);
2283         e1000_setup_rctl(adapter);
2284         e1000_configure_rx(adapter);
2285         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2286 }
2287
2288 /**
2289  * e1000e_power_up_phy - restore link in case the phy was powered down
2290  * @adapter: address of board private structure
2291  *
2292  * The phy may be powered down to save power and turn off link when the
2293  * driver is unloaded and wake on lan is not enabled (among others)
2294  * *** this routine MUST be followed by a call to e1000e_reset ***
2295  **/
2296 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2297 {
2298         u16 mii_reg = 0;
2299
2300         /* Just clear the power down bit to wake the phy back up */
2301         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2302                 /*
2303                  * According to the manual, the phy will retain its
2304                  * settings across a power-down/up cycle
2305                  */
2306                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2307                 mii_reg &= ~MII_CR_POWER_DOWN;
2308                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2309         }
2310
2311         adapter->hw.mac.ops.setup_link(&adapter->hw);
2312 }
2313
2314 /**
2315  * e1000_power_down_phy - Power down the PHY
2316  *
2317  * Power down the PHY so no link is implied when interface is down
2318  * The PHY cannot be powered down is management or WoL is active
2319  */
2320 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2321 {
2322         struct e1000_hw *hw = &adapter->hw;
2323         u16 mii_reg;
2324
2325         /* WoL is enabled */
2326         if (adapter->wol)
2327                 return;
2328
2329         /* non-copper PHY? */
2330         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2331                 return;
2332
2333         /* reset is blocked because of a SoL/IDER session */
2334         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2335                 return;
2336
2337         /* manageability (AMT) is enabled */
2338         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2339                 return;
2340
2341         /* power down the PHY */
2342         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2343         mii_reg |= MII_CR_POWER_DOWN;
2344         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2345         mdelay(1);
2346 }
2347
2348 /**
2349  * e1000e_reset - bring the hardware into a known good state
2350  *
2351  * This function boots the hardware and enables some settings that
2352  * require a configuration cycle of the hardware - those cannot be
2353  * set/changed during runtime. After reset the device needs to be
2354  * properly configured for Rx, Tx etc.
2355  */
2356 void e1000e_reset(struct e1000_adapter *adapter)
2357 {
2358         struct e1000_mac_info *mac = &adapter->hw.mac;
2359         struct e1000_fc_info *fc = &adapter->hw.fc;
2360         struct e1000_hw *hw = &adapter->hw;
2361         u32 tx_space, min_tx_space, min_rx_space;
2362         u32 pba = adapter->pba;
2363         u16 hwm;
2364
2365         /* reset Packet Buffer Allocation to default */
2366         ew32(PBA, pba);
2367
2368         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2369                 /*
2370                  * To maintain wire speed transmits, the Tx FIFO should be
2371                  * large enough to accommodate two full transmit packets,
2372                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2373                  * the Rx FIFO should be large enough to accommodate at least
2374                  * one full receive packet and is similarly rounded up and
2375                  * expressed in KB.
2376                  */
2377                 pba = er32(PBA);
2378                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2379                 tx_space = pba >> 16;
2380                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2381                 pba &= 0xffff;
2382                 /*
2383                  * the Tx fifo also stores 16 bytes of information about the tx
2384                  * but don't include ethernet FCS because hardware appends it
2385                  */
2386                 min_tx_space = (adapter->max_frame_size +
2387                                 sizeof(struct e1000_tx_desc) -
2388                                 ETH_FCS_LEN) * 2;
2389                 min_tx_space = ALIGN(min_tx_space, 1024);
2390                 min_tx_space >>= 10;
2391                 /* software strips receive CRC, so leave room for it */
2392                 min_rx_space = adapter->max_frame_size;
2393                 min_rx_space = ALIGN(min_rx_space, 1024);
2394                 min_rx_space >>= 10;
2395
2396                 /*
2397                  * If current Tx allocation is less than the min Tx FIFO size,
2398                  * and the min Tx FIFO size is less than the current Rx FIFO
2399                  * allocation, take space away from current Rx allocation
2400                  */
2401                 if ((tx_space < min_tx_space) &&
2402                     ((min_tx_space - tx_space) < pba)) {
2403                         pba -= min_tx_space - tx_space;
2404
2405                         /*
2406                          * if short on Rx space, Rx wins and must trump tx
2407                          * adjustment or use Early Receive if available
2408                          */
2409                         if ((pba < min_rx_space) &&
2410                             (!(adapter->flags & FLAG_HAS_ERT)))
2411                                 /* ERT enabled in e1000_configure_rx */
2412                                 pba = min_rx_space;
2413                 }
2414
2415                 ew32(PBA, pba);
2416         }
2417
2418
2419         /*
2420          * flow control settings
2421          *
2422          * The high water mark must be low enough to fit one full frame
2423          * (or the size used for early receive) above it in the Rx FIFO.
2424          * Set it to the lower of:
2425          * - 90% of the Rx FIFO size, and
2426          * - the full Rx FIFO size minus the early receive size (for parts
2427          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2428          * - the full Rx FIFO size minus one full frame
2429          */
2430         if (adapter->flags & FLAG_HAS_ERT)
2431                 hwm = min(((pba << 10) * 9 / 10),
2432                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2433         else
2434                 hwm = min(((pba << 10) * 9 / 10),
2435                           ((pba << 10) - adapter->max_frame_size));
2436
2437         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2438         fc->low_water = fc->high_water - 8;
2439
2440         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2441                 fc->pause_time = 0xFFFF;
2442         else
2443                 fc->pause_time = E1000_FC_PAUSE_TIME;
2444         fc->send_xon = 1;
2445         fc->type = fc->original_type;
2446
2447         /* Allow time for pending master requests to run */
2448         mac->ops.reset_hw(hw);
2449
2450         /*
2451          * For parts with AMT enabled, let the firmware know
2452          * that the network interface is in control
2453          */
2454         if ((adapter->flags & FLAG_HAS_AMT) && e1000e_check_mng_mode(hw))
2455                 e1000_get_hw_control(adapter);
2456
2457         ew32(WUC, 0);
2458
2459         if (mac->ops.init_hw(hw))
2460                 ndev_err(adapter->netdev, "Hardware Error\n");
2461
2462         e1000_update_mng_vlan(adapter);
2463
2464         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2465         ew32(VET, ETH_P_8021Q);
2466
2467         e1000e_reset_adaptive(hw);
2468         e1000_get_phy_info(hw);
2469
2470         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2471                 u16 phy_data = 0;
2472                 /*
2473                  * speed up time to link by disabling smart power down, ignore
2474                  * the return value of this function because there is nothing
2475                  * different we would do if it failed
2476                  */
2477                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2478                 phy_data &= ~IGP02E1000_PM_SPD;
2479                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2480         }
2481 }
2482
2483 int e1000e_up(struct e1000_adapter *adapter)
2484 {
2485         struct e1000_hw *hw = &adapter->hw;
2486
2487         /* hardware has been reset, we need to reload some things */
2488         e1000_configure(adapter);
2489
2490         clear_bit(__E1000_DOWN, &adapter->state);
2491
2492         napi_enable(&adapter->napi);
2493         e1000_irq_enable(adapter);
2494
2495         /* fire a link change interrupt to start the watchdog */
2496         ew32(ICS, E1000_ICS_LSC);
2497         return 0;
2498 }
2499
2500 void e1000e_down(struct e1000_adapter *adapter)
2501 {
2502         struct net_device *netdev = adapter->netdev;
2503         struct e1000_hw *hw = &adapter->hw;
2504         u32 tctl, rctl;
2505
2506         /*
2507          * signal that we're down so the interrupt handler does not
2508          * reschedule our watchdog timer
2509          */
2510         set_bit(__E1000_DOWN, &adapter->state);
2511
2512         /* disable receives in the hardware */
2513         rctl = er32(RCTL);
2514         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2515         /* flush and sleep below */
2516
2517         netif_stop_queue(netdev);
2518
2519         /* disable transmits in the hardware */
2520         tctl = er32(TCTL);
2521         tctl &= ~E1000_TCTL_EN;
2522         ew32(TCTL, tctl);
2523         /* flush both disables and wait for them to finish */
2524         e1e_flush();
2525         msleep(10);
2526
2527         napi_disable(&adapter->napi);
2528         e1000_irq_disable(adapter);
2529
2530         del_timer_sync(&adapter->watchdog_timer);
2531         del_timer_sync(&adapter->phy_info_timer);
2532
2533         netdev->tx_queue_len = adapter->tx_queue_len;
2534         netif_carrier_off(netdev);
2535         adapter->link_speed = 0;
2536         adapter->link_duplex = 0;
2537
2538         if (!pci_channel_offline(adapter->pdev))
2539                 e1000e_reset(adapter);
2540         e1000_clean_tx_ring(adapter);
2541         e1000_clean_rx_ring(adapter);
2542
2543         /*
2544          * TODO: for power management, we could drop the link and
2545          * pci_disable_device here.
2546          */
2547 }
2548
2549 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2550 {
2551         might_sleep();
2552         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2553                 msleep(1);
2554         e1000e_down(adapter);
2555         e1000e_up(adapter);
2556         clear_bit(__E1000_RESETTING, &adapter->state);
2557 }
2558
2559 /**
2560  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2561  * @adapter: board private structure to initialize
2562  *
2563  * e1000_sw_init initializes the Adapter private data structure.
2564  * Fields are initialized based on PCI device information and
2565  * OS network device settings (MTU size).
2566  **/
2567 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2568 {
2569         struct net_device *netdev = adapter->netdev;
2570
2571         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2572         adapter->rx_ps_bsize0 = 128;
2573         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2574         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2575
2576         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2577         if (!adapter->tx_ring)
2578                 goto err;
2579
2580         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2581         if (!adapter->rx_ring)
2582                 goto err;
2583
2584         spin_lock_init(&adapter->tx_queue_lock);
2585
2586         /* Explicitly disable IRQ since the NIC can be in any state. */
2587         e1000_irq_disable(adapter);
2588
2589         spin_lock_init(&adapter->stats_lock);
2590
2591         set_bit(__E1000_DOWN, &adapter->state);
2592         return 0;
2593
2594 err:
2595         ndev_err(netdev, "Unable to allocate memory for queues\n");
2596         kfree(adapter->rx_ring);
2597         kfree(adapter->tx_ring);
2598         return -ENOMEM;
2599 }
2600
2601 /**
2602  * e1000_open - Called when a network interface is made active
2603  * @netdev: network interface device structure
2604  *
2605  * Returns 0 on success, negative value on failure
2606  *
2607  * The open entry point is called when a network interface is made
2608  * active by the system (IFF_UP).  At this point all resources needed
2609  * for transmit and receive operations are allocated, the interrupt
2610  * handler is registered with the OS, the watchdog timer is started,
2611  * and the stack is notified that the interface is ready.
2612  **/
2613 static int e1000_open(struct net_device *netdev)
2614 {
2615         struct e1000_adapter *adapter = netdev_priv(netdev);
2616         struct e1000_hw *hw = &adapter->hw;
2617         int err;
2618
2619         /* disallow open during test */
2620         if (test_bit(__E1000_TESTING, &adapter->state))
2621                 return -EBUSY;
2622
2623         /* allocate transmit descriptors */
2624         err = e1000e_setup_tx_resources(adapter);
2625         if (err)
2626                 goto err_setup_tx;
2627
2628         /* allocate receive descriptors */
2629         err = e1000e_setup_rx_resources(adapter);
2630         if (err)
2631                 goto err_setup_rx;
2632
2633         e1000e_power_up_phy(adapter);
2634
2635         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2636         if ((adapter->hw.mng_cookie.status &
2637              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2638                 e1000_update_mng_vlan(adapter);
2639
2640         /*
2641          * If AMT is enabled, let the firmware know that the network
2642          * interface is now open
2643          */
2644         if ((adapter->flags & FLAG_HAS_AMT) &&
2645             e1000e_check_mng_mode(&adapter->hw))
2646                 e1000_get_hw_control(adapter);
2647
2648         /*
2649          * before we allocate an interrupt, we must be ready to handle it.
2650          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2651          * as soon as we call pci_request_irq, so we have to setup our
2652          * clean_rx handler before we do so.
2653          */
2654         e1000_configure(adapter);
2655
2656         err = e1000_request_irq(adapter);
2657         if (err)
2658                 goto err_req_irq;
2659
2660         /* From here on the code is the same as e1000e_up() */
2661         clear_bit(__E1000_DOWN, &adapter->state);
2662
2663         napi_enable(&adapter->napi);
2664
2665         e1000_irq_enable(adapter);
2666
2667         /* fire a link status change interrupt to start the watchdog */
2668         ew32(ICS, E1000_ICS_LSC);
2669
2670         return 0;
2671
2672 err_req_irq:
2673         e1000_release_hw_control(adapter);
2674         e1000_power_down_phy(adapter);
2675         e1000e_free_rx_resources(adapter);
2676 err_setup_rx:
2677         e1000e_free_tx_resources(adapter);
2678 err_setup_tx:
2679         e1000e_reset(adapter);
2680
2681         return err;
2682 }
2683
2684 /**
2685  * e1000_close - Disables a network interface
2686  * @netdev: network interface device structure
2687  *
2688  * Returns 0, this is not allowed to fail
2689  *
2690  * The close entry point is called when an interface is de-activated
2691  * by the OS.  The hardware is still under the drivers control, but
2692  * needs to be disabled.  A global MAC reset is issued to stop the
2693  * hardware, and all transmit and receive resources are freed.
2694  **/
2695 static int e1000_close(struct net_device *netdev)
2696 {
2697         struct e1000_adapter *adapter = netdev_priv(netdev);
2698
2699         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2700         e1000e_down(adapter);
2701         e1000_power_down_phy(adapter);
2702         e1000_free_irq(adapter);
2703
2704         e1000e_free_tx_resources(adapter);
2705         e1000e_free_rx_resources(adapter);
2706
2707         /*
2708          * kill manageability vlan ID if supported, but not if a vlan with
2709          * the same ID is registered on the host OS (let 8021q kill it)
2710          */
2711         if ((adapter->hw.mng_cookie.status &
2712                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2713              !(adapter->vlgrp &&
2714                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2715                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2716
2717         /*
2718          * If AMT is enabled, let the firmware know that the network
2719          * interface is now closed
2720          */
2721         if ((adapter->flags & FLAG_HAS_AMT) &&
2722             e1000e_check_mng_mode(&adapter->hw))
2723                 e1000_release_hw_control(adapter);
2724
2725         return 0;
2726 }
2727 /**
2728  * e1000_set_mac - Change the Ethernet Address of the NIC
2729  * @netdev: network interface device structure
2730  * @p: pointer to an address structure
2731  *
2732  * Returns 0 on success, negative on failure
2733  **/
2734 static int e1000_set_mac(struct net_device *netdev, void *p)
2735 {
2736         struct e1000_adapter *adapter = netdev_priv(netdev);
2737         struct sockaddr *addr = p;
2738
2739         if (!is_valid_ether_addr(addr->sa_data))
2740                 return -EADDRNOTAVAIL;
2741
2742         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2743         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2744
2745         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2746
2747         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2748                 /* activate the work around */
2749                 e1000e_set_laa_state_82571(&adapter->hw, 1);
2750
2751                 /*
2752                  * Hold a copy of the LAA in RAR[14] This is done so that
2753                  * between the time RAR[0] gets clobbered  and the time it
2754                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2755                  * of the RARs and no incoming packets directed to this port
2756                  * are dropped. Eventually the LAA will be in RAR[0] and
2757                  * RAR[14]
2758                  */
2759                 e1000e_rar_set(&adapter->hw,
2760                               adapter->hw.mac.addr,
2761                               adapter->hw.mac.rar_entry_count - 1);
2762         }
2763
2764         return 0;
2765 }
2766
2767 /*
2768  * Need to wait a few seconds after link up to get diagnostic information from
2769  * the phy
2770  */
2771 static void e1000_update_phy_info(unsigned long data)
2772 {
2773         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2774         e1000_get_phy_info(&adapter->hw);
2775 }
2776
2777 /**
2778  * e1000e_update_stats - Update the board statistics counters
2779  * @adapter: board private structure
2780  **/
2781 void e1000e_update_stats(struct e1000_adapter *adapter)
2782 {
2783         struct e1000_hw *hw = &adapter->hw;
2784         struct pci_dev *pdev = adapter->pdev;
2785         unsigned long irq_flags;
2786         u16 phy_tmp;
2787
2788 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2789
2790         /*
2791          * Prevent stats update while adapter is being reset, or if the pci
2792          * connection is down.
2793          */
2794         if (adapter->link_speed == 0)
2795                 return;
2796         if (pci_channel_offline(pdev))
2797                 return;
2798
2799         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2800
2801         /*
2802          * these counters are modified from e1000_adjust_tbi_stats,
2803          * called from the interrupt context, so they must only
2804          * be written while holding adapter->stats_lock
2805          */
2806
2807         adapter->stats.crcerrs += er32(CRCERRS);
2808         adapter->stats.gprc += er32(GPRC);
2809         adapter->stats.gorc += er32(GORCL);
2810         er32(GORCH); /* Clear gorc */
2811         adapter->stats.bprc += er32(BPRC);
2812         adapter->stats.mprc += er32(MPRC);
2813         adapter->stats.roc += er32(ROC);
2814
2815         adapter->stats.mpc += er32(MPC);
2816         adapter->stats.scc += er32(SCC);
2817         adapter->stats.ecol += er32(ECOL);
2818         adapter->stats.mcc += er32(MCC);
2819         adapter->stats.latecol += er32(LATECOL);
2820         adapter->stats.dc += er32(DC);
2821         adapter->stats.xonrxc += er32(XONRXC);
2822         adapter->stats.xontxc += er32(XONTXC);
2823         adapter->stats.xoffrxc += er32(XOFFRXC);
2824         adapter->stats.xofftxc += er32(XOFFTXC);
2825         adapter->stats.gptc += er32(GPTC);
2826         adapter->stats.gotc += er32(GOTCL);
2827         er32(GOTCH); /* Clear gotc */
2828         adapter->stats.rnbc += er32(RNBC);
2829         adapter->stats.ruc += er32(RUC);
2830
2831         adapter->stats.mptc += er32(MPTC);
2832         adapter->stats.bptc += er32(BPTC);
2833
2834         /* used for adaptive IFS */
2835
2836         hw->mac.tx_packet_delta = er32(TPT);
2837         adapter->stats.tpt += hw->mac.tx_packet_delta;
2838         hw->mac.collision_delta = er32(COLC);
2839         adapter->stats.colc += hw->mac.collision_delta;
2840
2841         adapter->stats.algnerrc += er32(ALGNERRC);
2842         adapter->stats.rxerrc += er32(RXERRC);
2843         adapter->stats.tncrs += er32(TNCRS);
2844         adapter->stats.cexterr += er32(CEXTERR);
2845         adapter->stats.tsctc += er32(TSCTC);
2846         adapter->stats.tsctfc += er32(TSCTFC);
2847
2848         /* Fill out the OS statistics structure */
2849         adapter->net_stats.multicast = adapter->stats.mprc;
2850         adapter->net_stats.collisions = adapter->stats.colc;
2851
2852         /* Rx Errors */
2853
2854         /*
2855          * RLEC on some newer hardware can be incorrect so build
2856          * our own version based on RUC and ROC
2857          */
2858         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2859                 adapter->stats.crcerrs + adapter->stats.algnerrc +
2860                 adapter->stats.ruc + adapter->stats.roc +
2861                 adapter->stats.cexterr;
2862         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2863                                               adapter->stats.roc;
2864         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2865         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2866         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2867
2868         /* Tx Errors */
2869         adapter->net_stats.tx_errors = adapter->stats.ecol +
2870                                        adapter->stats.latecol;
2871         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2872         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2873         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2874
2875         /* Tx Dropped needs to be maintained elsewhere */
2876
2877         /* Phy Stats */
2878         if (hw->phy.media_type == e1000_media_type_copper) {
2879                 if ((adapter->link_speed == SPEED_1000) &&
2880                    (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
2881                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2882                         adapter->phy_stats.idle_errors += phy_tmp;
2883                 }
2884         }
2885
2886         /* Management Stats */
2887         adapter->stats.mgptc += er32(MGTPTC);
2888         adapter->stats.mgprc += er32(MGTPRC);
2889         adapter->stats.mgpdc += er32(MGTPDC);
2890
2891         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2892 }
2893
2894 /**
2895  * e1000_phy_read_status - Update the PHY register status snapshot
2896  * @adapter: board private structure
2897  **/
2898 static void e1000_phy_read_status(struct e1000_adapter *adapter)
2899 {
2900         struct e1000_hw *hw = &adapter->hw;
2901         struct e1000_phy_regs *phy = &adapter->phy_regs;
2902         int ret_val;
2903         unsigned long irq_flags;
2904
2905
2906         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2907
2908         if ((er32(STATUS) & E1000_STATUS_LU) &&
2909             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
2910                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
2911                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
2912                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
2913                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
2914                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
2915                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
2916                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
2917                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
2918                 if (ret_val)
2919                         ndev_warn(adapter->netdev,
2920                                   "Error reading PHY register\n");
2921         } else {
2922                 /*
2923                  * Do not read PHY registers if link is not up
2924                  * Set values to typical power-on defaults
2925                  */
2926                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
2927                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
2928                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
2929                              BMSR_ERCAP);
2930                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
2931                                   ADVERTISE_ALL | ADVERTISE_CSMA);
2932                 phy->lpa = 0;
2933                 phy->expansion = EXPANSION_ENABLENPAGE;
2934                 phy->ctrl1000 = ADVERTISE_1000FULL;
2935                 phy->stat1000 = 0;
2936                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
2937         }
2938
2939         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2940 }
2941
2942 static void e1000_print_link_info(struct e1000_adapter *adapter)
2943 {
2944         struct e1000_hw *hw = &adapter->hw;
2945         struct net_device *netdev = adapter->netdev;
2946         u32 ctrl = er32(CTRL);
2947
2948         ndev_info(netdev,
2949                 "Link is Up %d Mbps %s, Flow Control: %s\n",
2950                 adapter->link_speed,
2951                 (adapter->link_duplex == FULL_DUPLEX) ?
2952                                 "Full Duplex" : "Half Duplex",
2953                 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
2954                                 "RX/TX" :
2955                 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
2956                 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
2957 }
2958
2959 static bool e1000_has_link(struct e1000_adapter *adapter)
2960 {
2961         struct e1000_hw *hw = &adapter->hw;
2962         bool link_active = 0;
2963         s32 ret_val = 0;
2964
2965         /*
2966          * get_link_status is set on LSC (link status) interrupt or
2967          * Rx sequence error interrupt.  get_link_status will stay
2968          * false until the check_for_link establishes link
2969          * for copper adapters ONLY
2970          */
2971         switch (hw->phy.media_type) {
2972         case e1000_media_type_copper:
2973                 if (hw->mac.get_link_status) {
2974                         ret_val = hw->mac.ops.check_for_link(hw);
2975                         link_active = !hw->mac.get_link_status;
2976                 } else {
2977                         link_active = 1;
2978                 }
2979                 break;
2980         case e1000_media_type_fiber:
2981                 ret_val = hw->mac.ops.check_for_link(hw);
2982                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2983                 break;
2984         case e1000_media_type_internal_serdes:
2985                 ret_val = hw->mac.ops.check_for_link(hw);
2986                 link_active = adapter->hw.mac.serdes_has_link;
2987                 break;
2988         default:
2989         case e1000_media_type_unknown:
2990                 break;
2991         }
2992
2993         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
2994             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
2995                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
2996                 ndev_info(adapter->netdev,
2997                           "Gigabit has been disabled, downgrading speed\n");
2998         }
2999
3000         return link_active;
3001 }
3002
3003 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3004 {
3005         /* make sure the receive unit is started */
3006         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3007             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3008                 struct e1000_hw *hw = &adapter->hw;
3009                 u32 rctl = er32(RCTL);
3010                 ew32(RCTL, rctl | E1000_RCTL_EN);
3011                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3012         }
3013 }
3014
3015 /**
3016  * e1000_watchdog - Timer Call-back
3017  * @data: pointer to adapter cast into an unsigned long
3018  **/
3019 static void e1000_watchdog(unsigned long data)
3020 {
3021         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3022
3023         /* Do the rest outside of interrupt context */
3024         schedule_work(&adapter->watchdog_task);
3025
3026         /* TODO: make this use queue_delayed_work() */
3027 }
3028
3029 static void e1000_watchdog_task(struct work_struct *work)
3030 {
3031         struct e1000_adapter *adapter = container_of(work,
3032                                         struct e1000_adapter, watchdog_task);
3033         struct net_device *netdev = adapter->netdev;
3034         struct e1000_mac_info *mac = &adapter->hw.mac;
3035         struct e1000_ring *tx_ring = adapter->tx_ring;
3036         struct e1000_hw *hw = &adapter->hw;
3037         u32 link, tctl;
3038         int tx_pending = 0;
3039
3040         link = e1000_has_link(adapter);
3041         if ((netif_carrier_ok(netdev)) && link) {
3042                 e1000e_enable_receives(adapter);
3043                 goto link_up;
3044         }
3045
3046         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3047             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3048                 e1000_update_mng_vlan(adapter);
3049
3050         if (link) {
3051                 if (!netif_carrier_ok(netdev)) {
3052                         bool txb2b = 1;
3053                         /* update snapshot of PHY registers on LSC */
3054                         e1000_phy_read_status(adapter);
3055                         mac->ops.get_link_up_info(&adapter->hw,
3056                                                    &adapter->link_speed,
3057                                                    &adapter->link_duplex);
3058                         e1000_print_link_info(adapter);
3059                         /*
3060                          * tweak tx_queue_len according to speed/duplex
3061                          * and adjust the timeout factor
3062                          */
3063                         netdev->tx_queue_len = adapter->tx_queue_len;
3064                         adapter->tx_timeout_factor = 1;
3065                         switch (adapter->link_speed) {
3066                         case SPEED_10:
3067                                 txb2b = 0;
3068                                 netdev->tx_queue_len = 10;
3069                                 adapter->tx_timeout_factor = 14;
3070                                 break;
3071                         case SPEED_100:
3072                                 txb2b = 0;
3073                                 netdev->tx_queue_len = 100;
3074                                 /* maybe add some timeout factor ? */
3075                                 break;
3076                         }
3077
3078                         /*
3079                          * workaround: re-program speed mode bit after
3080                          * link-up event
3081                          */
3082                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3083                             !txb2b) {
3084                                 u32 tarc0;
3085                                 tarc0 = er32(TARC(0));
3086                                 tarc0 &= ~SPEED_MODE_BIT;
3087                                 ew32(TARC(0), tarc0);
3088                         }
3089
3090                         /*
3091                          * disable TSO for pcie and 10/100 speeds, to avoid
3092                          * some hardware issues
3093                          */
3094                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3095                                 switch (adapter->link_speed) {
3096                                 case SPEED_10:
3097                                 case SPEED_100:
3098                                         ndev_info(netdev,
3099                                         "10/100 speed: disabling TSO\n");
3100                                         netdev->features &= ~NETIF_F_TSO;
3101                                         netdev->features &= ~NETIF_F_TSO6;
3102                                         break;
3103                                 case SPEED_1000:
3104                                         netdev->features |= NETIF_F_TSO;
3105                                         netdev->features |= NETIF_F_TSO6;
3106                                         break;
3107                                 default:
3108                                         /* oops */
3109                                         break;
3110                                 }
3111                         }
3112
3113                         /*
3114                          * enable transmits in the hardware, need to do this
3115                          * after setting TARC(0)
3116                          */
3117                         tctl = er32(TCTL);
3118                         tctl |= E1000_TCTL_EN;
3119                         ew32(TCTL, tctl);
3120
3121                         netif_carrier_on(netdev);
3122                         netif_wake_queue(netdev);
3123
3124                         if (!test_bit(__E1000_DOWN, &adapter->state))
3125                                 mod_timer(&adapter->phy_info_timer,
3126                                           round_jiffies(jiffies + 2 * HZ));
3127                 }
3128         } else {
3129                 if (netif_carrier_ok(netdev)) {
3130                         adapter->link_speed = 0;
3131                         adapter->link_duplex = 0;
3132                         ndev_info(netdev, "Link is Down\n");
3133                         netif_carrier_off(netdev);
3134                         netif_stop_queue(netdev);
3135                         if (!test_bit(__E1000_DOWN, &adapter->state))
3136                                 mod_timer(&adapter->phy_info_timer,
3137                                           round_jiffies(jiffies + 2 * HZ));
3138
3139                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3140                                 schedule_work(&adapter->reset_task);
3141                 }
3142         }
3143
3144 link_up:
3145         e1000e_update_stats(adapter);
3146
3147         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3148         adapter->tpt_old = adapter->stats.tpt;
3149         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3150         adapter->colc_old = adapter->stats.colc;
3151
3152         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3153         adapter->gorc_old = adapter->stats.gorc;
3154         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3155         adapter->gotc_old = adapter->stats.gotc;
3156
3157         e1000e_update_adaptive(&adapter->hw);
3158
3159         if (!netif_carrier_ok(netdev)) {
3160                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3161                                tx_ring->count);
3162                 if (tx_pending) {
3163                         /*
3164                          * We've lost link, so the controller stops DMA,
3165                          * but we've got queued Tx work that's never going
3166                          * to get done, so reset controller to flush Tx.
3167                          * (Do the reset outside of interrupt context).
3168                          */
3169                         adapter->tx_timeout_count++;
3170                         schedule_work(&adapter->reset_task);
3171                 }
3172         }
3173
3174         /* Cause software interrupt to ensure Rx ring is cleaned */
3175         ew32(ICS, E1000_ICS_RXDMT0);
3176
3177         /* Force detection of hung controller every watchdog period */
3178         adapter->detect_tx_hung = 1;
3179
3180         /*
3181          * With 82571 controllers, LAA may be overwritten due to controller
3182          * reset from the other port. Set the appropriate LAA in RAR[0]
3183          */
3184         if (e1000e_get_laa_state_82571(hw))
3185                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3186
3187         /* Reset the timer */
3188         if (!test_bit(__E1000_DOWN, &adapter->state))
3189                 mod_timer(&adapter->watchdog_timer,
3190                           round_jiffies(jiffies + 2 * HZ));
3191 }
3192
3193 #define E1000_TX_FLAGS_CSUM             0x00000001
3194 #define E1000_TX_FLAGS_VLAN             0x00000002
3195 #define E1000_TX_FLAGS_TSO              0x00000004
3196 #define E1000_TX_FLAGS_IPV4             0x00000008
3197 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3198 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3199
3200 static int e1000_tso(struct e1000_adapter *adapter,
3201                      struct sk_buff *skb)
3202 {
3203         struct e1000_ring *tx_ring = adapter->tx_ring;
3204         struct e1000_context_desc *context_desc;
3205         struct e1000_buffer *buffer_info;
3206         unsigned int i;
3207         u32 cmd_length = 0;
3208         u16 ipcse = 0, tucse, mss;
3209         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3210         int err;
3211
3212         if (skb_is_gso(skb)) {
3213                 if (skb_header_cloned(skb)) {
3214                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3215                         if (err)
3216                                 return err;
3217                 }
3218
3219                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3220                 mss = skb_shinfo(skb)->gso_size;
3221                 if (skb->protocol == htons(ETH_P_IP)) {
3222                         struct iphdr *iph = ip_hdr(skb);
3223                         iph->tot_len = 0;
3224                         iph->check = 0;
3225                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3226                                                                  iph->daddr, 0,
3227                                                                  IPPROTO_TCP,
3228                                                                  0);
3229                         cmd_length = E1000_TXD_CMD_IP;
3230                         ipcse = skb_transport_offset(skb) - 1;
3231                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3232                         ipv6_hdr(skb)->payload_len = 0;
3233                         tcp_hdr(skb)->check =
3234                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3235                                                  &ipv6_hdr(skb)->daddr,
3236                                                  0, IPPROTO_TCP, 0);
3237                         ipcse = 0;
3238                 }
3239                 ipcss = skb_network_offset(skb);
3240                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3241                 tucss = skb_transport_offset(skb);
3242                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3243                 tucse = 0;
3244
3245                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3246                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3247
3248                 i = tx_ring->next_to_use;
3249                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3250                 buffer_info = &tx_ring->buffer_info[i];
3251
3252                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3253                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3254                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3255                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3256                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3257                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3258                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3259                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3260                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3261
3262                 buffer_info->time_stamp = jiffies;
3263                 buffer_info->next_to_watch = i;
3264
3265                 i++;
3266                 if (i == tx_ring->count)
3267                         i = 0;
3268                 tx_ring->next_to_use = i;
3269
3270                 return 1;
3271         }
3272
3273         return 0;
3274 }
3275
3276 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3277 {
3278         struct e1000_ring *tx_ring = adapter->tx_ring;
3279         struct e1000_context_desc *context_desc;
3280         struct e1000_buffer *buffer_info;
3281         unsigned int i;
3282         u8 css;
3283
3284         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3285                 css = skb_transport_offset(skb);
3286
3287                 i = tx_ring->next_to_use;
3288                 buffer_info = &tx_ring->buffer_info[i];
3289                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3290
3291                 context_desc->lower_setup.ip_config = 0;
3292                 context_desc->upper_setup.tcp_fields.tucss = css;
3293                 context_desc->upper_setup.tcp_fields.tucso =
3294                                         css + skb->csum_offset;
3295                 context_desc->upper_setup.tcp_fields.tucse = 0;
3296                 context_desc->tcp_seg_setup.data = 0;
3297                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
3298
3299                 buffer_info->time_stamp = jiffies;
3300                 buffer_info->next_to_watch = i;
3301
3302                 i++;
3303                 if (i == tx_ring->count)
3304                         i = 0;
3305                 tx_ring->next_to_use = i;
3306
3307                 return 1;
3308         }
3309
3310         return 0;
3311 }
3312
3313 #define E1000_MAX_PER_TXD       8192
3314 #define E1000_MAX_TXD_PWR       12
3315
3316 static int e1000_tx_map(struct e1000_adapter *adapter,
3317                         struct sk_buff *skb, unsigned int first,
3318                         unsigned int max_per_txd, unsigned int nr_frags,
3319                         unsigned int mss)
3320 {
3321         struct e1000_ring *tx_ring = adapter->tx_ring;
3322         struct e1000_buffer *buffer_info;
3323         unsigned int len = skb->len - skb->data_len;
3324         unsigned int offset = 0, size, count = 0, i;
3325         unsigned int f;
3326
3327         i = tx_ring->next_to_use;
3328
3329         while (len) {
3330                 buffer_info = &tx_ring->buffer_info[i];
3331                 size = min(len, max_per_txd);
3332
3333                 /* Workaround for premature desc write-backs
3334                  * in TSO mode.  Append 4-byte sentinel desc */
3335                 if (mss && !nr_frags && size == len && size > 8)
3336                         size -= 4;
3337
3338                 buffer_info->length = size;
3339                 /* set time_stamp *before* dma to help avoid a possible race */
3340                 buffer_info->time_stamp = jiffies;
3341                 buffer_info->dma =
3342                         pci_map_single(adapter->pdev,
3343                                 skb->data + offset,
3344                                 size,
3345                                 PCI_DMA_TODEVICE);
3346                 if (pci_dma_mapping_error(buffer_info->dma)) {
3347                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3348                         adapter->tx_dma_failed++;
3349                         return -1;
3350                 }
3351                 buffer_info->next_to_watch = i;
3352
3353                 len -= size;
3354                 offset += size;
3355                 count++;
3356                 i++;
3357                 if (i == tx_ring->count)
3358                         i = 0;
3359         }
3360
3361         for (f = 0; f < nr_frags; f++) {
3362                 struct skb_frag_struct *frag;
3363
3364                 frag = &skb_shinfo(skb)->frags[f];
3365                 len = frag->size;
3366                 offset = frag->page_offset;
3367
3368                 while (len) {
3369                         buffer_info = &tx_ring->buffer_info[i];
3370                         size = min(len, max_per_txd);
3371                         /* Workaround for premature desc write-backs
3372                          * in TSO mode.  Append 4-byte sentinel desc */
3373                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3374                                 size -= 4;
3375
3376                         buffer_info->length = size;
3377                         buffer_info->time_stamp = jiffies;
3378                         buffer_info->dma =
3379                                 pci_map_page(adapter->pdev,
3380                                         frag->page,
3381                                         offset,
3382                                         size,
3383                                         PCI_DMA_TODEVICE);
3384                         if (pci_dma_mapping_error(buffer_info->dma)) {
3385                                 dev_err(&adapter->pdev->dev,
3386                                         "TX DMA page map failed\n");
3387                                 adapter->tx_dma_failed++;
3388                                 return -1;
3389                         }
3390
3391                         buffer_info->next_to_watch = i;
3392
3393                         len -= size;
3394                         offset += size;
3395                         count++;
3396
3397                         i++;
3398                         if (i == tx_ring->count)
3399                                 i = 0;
3400                 }
3401         }
3402
3403         if (i == 0)
3404                 i = tx_ring->count - 1;
3405         else
3406                 i--;
3407
3408         tx_ring->buffer_info[i].skb = skb;
3409         tx_ring->buffer_info[first].next_to_watch = i;
3410
3411         return count;
3412 }
3413
3414 static void e1000_tx_queue(struct e1000_adapter *adapter,
3415                            int tx_flags, int count)
3416 {
3417         struct e1000_ring *tx_ring = adapter->tx_ring;
3418         struct e1000_tx_desc *tx_desc = NULL;
3419         struct e1000_buffer *buffer_info;
3420         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3421         unsigned int i;
3422
3423         if (tx_flags & E1000_TX_FLAGS_TSO) {
3424                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3425                              E1000_TXD_CMD_TSE;
3426                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3427
3428                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3429                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3430         }
3431
3432         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3433                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3434                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3435         }
3436
3437         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3438                 txd_lower |= E1000_TXD_CMD_VLE;
3439                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3440         }
3441
3442         i = tx_ring->next_to_use;
3443
3444         while (count--) {
3445                 buffer_info = &tx_ring->buffer_info[i];
3446                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3447                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3448                 tx_desc->lower.data =
3449                         cpu_to_le32(txd_lower | buffer_info->length);
3450                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3451
3452                 i++;
3453                 if (i == tx_ring->count)
3454                         i = 0;
3455         }
3456
3457         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3458
3459         /*
3460          * Force memory writes to complete before letting h/w
3461          * know there are new descriptors to fetch.  (Only
3462          * applicable for weak-ordered memory model archs,
3463          * such as IA-64).
3464          */
3465         wmb();
3466
3467         tx_ring->next_to_use = i;
3468         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3469         /*
3470          * we need this if more than one processor can write to our tail
3471          * at a time, it synchronizes IO on IA64/Altix systems
3472          */
3473         mmiowb();
3474 }
3475
3476 #define MINIMUM_DHCP_PACKET_SIZE 282
3477 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3478                                     struct sk_buff *skb)
3479 {
3480         struct e1000_hw *hw =  &adapter->hw;
3481         u16 length, offset;
3482
3483         if (vlan_tx_tag_present(skb)) {
3484                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3485                     && (adapter->hw.mng_cookie.status &
3486                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3487                         return 0;
3488         }
3489
3490         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3491                 return 0;
3492
3493         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3494                 return 0;
3495
3496         {
3497                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3498                 struct udphdr *udp;
3499
3500                 if (ip->protocol != IPPROTO_UDP)
3501                         return 0;
3502
3503                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3504                 if (ntohs(udp->dest) != 67)
3505                         return 0;
3506
3507                 offset = (u8 *)udp + 8 - skb->data;
3508                 length = skb->len - offset;
3509                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3510         }
3511
3512         return 0;
3513 }
3514
3515 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3516 {
3517         struct e1000_adapter *adapter = netdev_priv(netdev);
3518
3519         netif_stop_queue(netdev);
3520         /*
3521          * Herbert's original patch had:
3522          *  smp_mb__after_netif_stop_queue();
3523          * but since that doesn't exist yet, just open code it.
3524          */
3525         smp_mb();
3526
3527         /*
3528          * We need to check again in a case another CPU has just
3529          * made room available.
3530          */
3531         if (e1000_desc_unused(adapter->tx_ring) < size)
3532                 return -EBUSY;
3533
3534         /* A reprieve! */
3535         netif_start_queue(netdev);
3536         ++adapter->restart_queue;
3537         return 0;
3538 }
3539
3540 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3541 {
3542         struct e1000_adapter *adapter = netdev_priv(netdev);
3543
3544         if (e1000_desc_unused(adapter->tx_ring) >= size)
3545                 return 0;
3546         return __e1000_maybe_stop_tx(netdev, size);
3547 }
3548
3549 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3550 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3551 {
3552         struct e1000_adapter *adapter = netdev_priv(netdev);
3553         struct e1000_ring *tx_ring = adapter->tx_ring;
3554         unsigned int first;
3555         unsigned int max_per_txd = E1000_MAX_PER_TXD;
3556         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3557         unsigned int tx_flags = 0;
3558         unsigned int len = skb->len - skb->data_len;
3559         unsigned long irq_flags;
3560         unsigned int nr_frags;
3561         unsigned int mss;
3562         int count = 0;
3563         int tso;
3564         unsigned int f;
3565
3566         if (test_bit(__E1000_DOWN, &adapter->state)) {
3567                 dev_kfree_skb_any(skb);
3568                 return NETDEV_TX_OK;
3569         }
3570
3571         if (skb->len <= 0) {
3572                 dev_kfree_skb_any(skb);
3573                 return NETDEV_TX_OK;
3574         }
3575
3576         mss = skb_shinfo(skb)->gso_size;
3577         /*
3578          * The controller does a simple calculation to
3579          * make sure there is enough room in the FIFO before
3580          * initiating the DMA for each buffer.  The calc is:
3581          * 4 = ceil(buffer len/mss).  To make sure we don't
3582          * overrun the FIFO, adjust the max buffer len if mss
3583          * drops.
3584          */
3585         if (mss) {
3586                 u8 hdr_len;
3587                 max_per_txd = min(mss << 2, max_per_txd);
3588                 max_txd_pwr = fls(max_per_txd) - 1;
3589
3590                 /*
3591                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
3592                  * points to just header, pull a few bytes of payload from
3593                  * frags into skb->data
3594                  */
3595                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3596                 /*
3597                  * we do this workaround for ES2LAN, but it is un-necessary,
3598                  * avoiding it could save a lot of cycles
3599                  */
3600                 if (skb->data_len && (hdr_len == len)) {
3601                         unsigned int pull_size;
3602
3603                         pull_size = min((unsigned int)4, skb->data_len);
3604                         if (!__pskb_pull_tail(skb, pull_size)) {
3605                                 ndev_err(netdev,
3606                                          "__pskb_pull_tail failed.\n");
3607                                 dev_kfree_skb_any(skb);
3608                                 return NETDEV_TX_OK;
3609                         }
3610                         len = skb->len - skb->data_len;
3611                 }
3612         }
3613
3614         /* reserve a descriptor for the offload context */
3615         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3616                 count++;
3617         count++;
3618
3619         count += TXD_USE_COUNT(len, max_txd_pwr);
3620
3621         nr_frags = skb_shinfo(skb)->nr_frags;
3622         for (f = 0; f < nr_frags; f++)
3623                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3624                                        max_txd_pwr);
3625
3626         if (adapter->hw.mac.tx_pkt_filtering)
3627                 e1000_transfer_dhcp_info(adapter, skb);
3628
3629         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3630                 /* Collision - tell upper layer to requeue */
3631                 return NETDEV_TX_LOCKED;
3632
3633         /*
3634          * need: count + 2 desc gap to keep tail from touching
3635          * head, otherwise try next time
3636          */
3637         if (e1000_maybe_stop_tx(netdev, count + 2)) {
3638                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3639                 return NETDEV_TX_BUSY;
3640         }
3641
3642         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3643                 tx_flags |= E1000_TX_FLAGS_VLAN;
3644                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3645         }
3646
3647         first = tx_ring->next_to_use;
3648
3649         tso = e1000_tso(adapter, skb);
3650         if (tso < 0) {
3651                 dev_kfree_skb_any(skb);
3652                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3653                 return NETDEV_TX_OK;
3654         }
3655
3656         if (tso)
3657                 tx_flags |= E1000_TX_FLAGS_TSO;
3658         else if (e1000_tx_csum(adapter, skb))
3659                 tx_flags |= E1000_TX_FLAGS_CSUM;
3660
3661         /*
3662          * Old method was to assume IPv4 packet by default if TSO was enabled.
3663          * 82571 hardware supports TSO capabilities for IPv6 as well...
3664          * no longer assume, we must.
3665          */
3666         if (skb->protocol == htons(ETH_P_IP))
3667                 tx_flags |= E1000_TX_FLAGS_IPV4;
3668
3669         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3670         if (count < 0) {
3671                 /* handle pci_map_single() error in e1000_tx_map */
3672                 dev_kfree_skb_any(skb);
3673                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3674                 return NETDEV_TX_OK;
3675         }
3676
3677         e1000_tx_queue(adapter, tx_flags, count);
3678
3679         netdev->trans_start = jiffies;
3680
3681         /* Make sure there is space in the ring for the next send. */
3682         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3683
3684         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3685         return NETDEV_TX_OK;
3686 }
3687
3688 /**
3689  * e1000_tx_timeout - Respond to a Tx Hang
3690  * @netdev: network interface device structure
3691  **/
3692 static void e1000_tx_timeout(struct net_device *netdev)
3693 {
3694         struct e1000_adapter *adapter = netdev_priv(netdev);
3695
3696         /* Do the reset outside of interrupt context */
3697         adapter->tx_timeout_count++;
3698         schedule_work(&adapter->reset_task);
3699 }
3700
3701 static void e1000_reset_task(struct work_struct *work)
3702 {
3703         struct e1000_adapter *adapter;
3704         adapter = container_of(work, struct e1000_adapter, reset_task);
3705
3706         e1000e_reinit_locked(adapter);
3707 }
3708
3709 /**
3710  * e1000_get_stats - Get System Network Statistics
3711  * @netdev: network interface device structure
3712  *
3713  * Returns the address of the device statistics structure.
3714  * The statistics are actually updated from the timer callback.
3715  **/
3716 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3717 {
3718         struct e1000_adapter *adapter = netdev_priv(netdev);
3719
3720         /* only return the current stats */
3721         return &adapter->net_stats;
3722 }
3723
3724 /**
3725  * e1000_change_mtu - Change the Maximum Transfer Unit
3726  * @netdev: network interface device structure
3727  * @new_mtu: new value for maximum frame size
3728  *
3729  * Returns 0 on success, negative on failure
3730  **/
3731 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3732 {
3733         struct e1000_adapter *adapter = netdev_priv(netdev);
3734         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3735
3736         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3737             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3738                 ndev_err(netdev, "Invalid MTU setting\n");
3739                 return -EINVAL;
3740         }
3741
3742         /* Jumbo frame size limits */
3743         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3744                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3745                         ndev_err(netdev, "Jumbo Frames not supported.\n");
3746                         return -EINVAL;
3747                 }
3748                 if (adapter->hw.phy.type == e1000_phy_ife) {
3749                         ndev_err(netdev, "Jumbo Frames not supported.\n");
3750                         return -EINVAL;
3751                 }
3752         }
3753
3754 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3755         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3756                 ndev_err(netdev, "MTU > 9216 not supported.\n");
3757                 return -EINVAL;
3758         }
3759
3760         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3761                 msleep(1);
3762         /* e1000e_down has a dependency on max_frame_size */
3763         adapter->max_frame_size = max_frame;
3764         if (netif_running(netdev))
3765                 e1000e_down(adapter);
3766
3767         /*
3768          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3769          * means we reserve 2 more, this pushes us to allocate from the next
3770          * larger slab size.
3771          * i.e. RXBUFFER_2048 --> size-4096 slab
3772          * However with the new *_jumbo_rx* routines, jumbo receives will use
3773          * fragmented skbs
3774          */
3775
3776         if (max_frame <= 256)
3777                 adapter->rx_buffer_len = 256;
3778         else if (max_frame <= 512)
3779                 adapter->rx_buffer_len = 512;
3780         else if (max_frame <= 1024)
3781                 adapter->rx_buffer_len = 1024;
3782         else if (max_frame <= 2048)
3783                 adapter->rx_buffer_len = 2048;
3784         else
3785                 adapter->rx_buffer_len = 4096;
3786
3787         /* adjust allocation if LPE protects us, and we aren't using SBP */
3788         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3789              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3790                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3791                                          + ETH_FCS_LEN;
3792
3793         ndev_info(netdev, "changing MTU from %d to %d\n",
3794                 netdev->mtu, new_mtu);
3795         netdev->mtu = new_mtu;
3796
3797         if (netif_running(netdev))
3798                 e1000e_up(adapter);
3799         else
3800                 e1000e_reset(adapter);
3801
3802         clear_bit(__E1000_RESETTING, &adapter->state);
3803
3804         return 0;
3805 }
3806
3807 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3808                            int cmd)
3809 {
3810         struct e1000_adapter *adapter = netdev_priv(netdev);
3811         struct mii_ioctl_data *data = if_mii(ifr);
3812
3813         if (adapter->hw.phy.media_type != e1000_media_type_copper)
3814                 return -EOPNOTSUPP;
3815
3816         switch (cmd) {
3817         case SIOCGMIIPHY:
3818                 data->phy_id = adapter->hw.phy.addr;
3819                 break;
3820         case SIOCGMIIREG:
3821                 if (!capable(CAP_NET_ADMIN))
3822                         return -EPERM;
3823                 switch (data->reg_num & 0x1F) {
3824                 case MII_BMCR:
3825                         data->val_out = adapter->phy_regs.bmcr;
3826                         break;
3827                 case MII_BMSR:
3828                         data->val_out = adapter->phy_regs.bmsr;
3829                         break;
3830                 case MII_PHYSID1:
3831                         data->val_out = (adapter->hw.phy.id >> 16);
3832                         break;
3833                 case MII_PHYSID2:
3834                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
3835                         break;
3836                 case MII_ADVERTISE:
3837                         data->val_out = adapter->phy_regs.advertise;
3838                         break;
3839                 case MII_LPA:
3840                         data->val_out = adapter->phy_regs.lpa;
3841                         break;
3842                 case MII_EXPANSION:
3843                         data->val_out = adapter->phy_regs.expansion;
3844                         break;
3845                 case MII_CTRL1000:
3846                         data->val_out = adapter->phy_regs.ctrl1000;
3847                         break;
3848                 case MII_STAT1000:
3849                         data->val_out = adapter->phy_regs.stat1000;
3850                         break;
3851                 case MII_ESTATUS:
3852                         data->val_out = adapter->phy_regs.estatus;
3853                         break;
3854                 default:
3855                         return -EIO;
3856                 }
3857                 break;
3858         case SIOCSMIIREG:
3859         default:
3860                 return -EOPNOTSUPP;
3861         }
3862         return 0;
3863 }
3864
3865 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3866 {
3867         switch (cmd) {
3868         case SIOCGMIIPHY:
3869         case SIOCGMIIREG:
3870         case SIOCSMIIREG:
3871                 return e1000_mii_ioctl(netdev, ifr, cmd);
3872         default:
3873                 return -EOPNOTSUPP;
3874         }
3875 }
3876
3877 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
3878 {
3879         struct net_device *netdev = pci_get_drvdata(pdev);
3880         struct e1000_adapter *adapter = netdev_priv(netdev);
3881         struct e1000_hw *hw = &adapter->hw;
3882         u32 ctrl, ctrl_ext, rctl, status;
3883         u32 wufc = adapter->wol;
3884         int retval = 0;
3885
3886         netif_device_detach(netdev);
3887
3888         if (netif_running(netdev)) {
3889                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3890                 e1000e_down(adapter);
3891                 e1000_free_irq(adapter);
3892         }
3893
3894         retval = pci_save_state(pdev);
3895         if (retval)
3896                 return retval;
3897
3898         status = er32(STATUS);
3899         if (status & E1000_STATUS_LU)
3900                 wufc &= ~E1000_WUFC_LNKC;
3901
3902         if (wufc) {
3903                 e1000_setup_rctl(adapter);
3904                 e1000_set_multi(netdev);
3905
3906                 /* turn on all-multi mode if wake on multicast is enabled */
3907                 if (wufc & E1000_WUFC_MC) {
3908                         rctl = er32(RCTL);
3909                         rctl |= E1000_RCTL_MPE;
3910                         ew32(RCTL, rctl);
3911                 }
3912
3913                 ctrl = er32(CTRL);
3914                 /* advertise wake from D3Cold */
3915                 #define E1000_CTRL_ADVD3WUC 0x00100000
3916                 /* phy power management enable */
3917                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3918                 ctrl |= E1000_CTRL_ADVD3WUC |
3919                         E1000_CTRL_EN_PHY_PWR_MGMT;
3920                 ew32(CTRL, ctrl);
3921
3922                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
3923                     adapter->hw.phy.media_type ==
3924                     e1000_media_type_internal_serdes) {
3925                         /* keep the laser running in D3 */
3926                         ctrl_ext = er32(CTRL_EXT);
3927                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3928                         ew32(CTRL_EXT, ctrl_ext);
3929                 }
3930
3931                 if (adapter->flags & FLAG_IS_ICH)
3932                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
3933
3934                 /* Allow time for pending master requests to run */
3935                 e1000e_disable_pcie_master(&adapter->hw);
3936
3937                 ew32(WUC, E1000_WUC_PME_EN);
3938                 ew32(WUFC, wufc);
3939                 pci_enable_wake(pdev, PCI_D3hot, 1);
3940                 pci_enable_wake(pdev, PCI_D3cold, 1);
3941         } else {
3942                 ew32(WUC, 0);
3943                 ew32(WUFC, 0);
3944                 pci_enable_wake(pdev, PCI_D3hot, 0);
3945                 pci_enable_wake(pdev, PCI_D3cold, 0);
3946         }
3947
3948         /* make sure adapter isn't asleep if manageability is enabled */
3949         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
3950                 pci_enable_wake(pdev, PCI_D3hot, 1);
3951                 pci_enable_wake(pdev, PCI_D3cold, 1);
3952         }
3953
3954         if (adapter->hw.phy.type == e1000_phy_igp_3)
3955                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
3956
3957         /*
3958          * Release control of h/w to f/w.  If f/w is AMT enabled, this
3959          * would have already happened in close and is redundant.
3960          */
3961         e1000_release_hw_control(adapter);
3962
3963         pci_disable_device(pdev);
3964
3965         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3966
3967         return 0;
3968 }
3969
3970 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
3971 {
3972         int pos;
3973         u16 val;
3974
3975         /*
3976          * 82573 workaround - disable L1 ASPM on mobile chipsets
3977          *
3978          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
3979          * resulting in lost data or garbage information on the pci-e link
3980          * level. This could result in (false) bad EEPROM checksum errors,
3981          * long ping times (up to 2s) or even a system freeze/hang.
3982          *
3983          * Unfortunately this feature saves about 1W power consumption when
3984          * active.
3985          */
3986         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
3987         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
3988         if (val & 0x2) {
3989                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
3990                 val &= ~0x2;
3991                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
3992         }
3993 }
3994
3995 #ifdef CONFIG_PM
3996 static int e1000_resume(struct pci_dev *pdev)
3997 {
3998         struct net_device *netdev = pci_get_drvdata(pdev);
3999         struct e1000_adapter *adapter = netdev_priv(netdev);
4000         struct e1000_hw *hw = &adapter->hw;
4001         u32 err;
4002
4003         pci_set_power_state(pdev, PCI_D0);
4004         pci_restore_state(pdev);
4005         e1000e_disable_l1aspm(pdev);
4006         err = pci_enable_device(pdev);
4007         if (err) {
4008                 dev_err(&pdev->dev,
4009                         "Cannot enable PCI device from suspend\n");
4010                 return err;
4011         }
4012
4013         pci_set_master(pdev);
4014
4015         pci_enable_wake(pdev, PCI_D3hot, 0);
4016         pci_enable_wake(pdev, PCI_D3cold, 0);
4017
4018         if (netif_running(netdev)) {
4019                 err = e1000_request_irq(adapter);
4020                 if (err)
4021                         return err;
4022         }
4023
4024         e1000e_power_up_phy(adapter);
4025         e1000e_reset(adapter);
4026         ew32(WUS, ~0);
4027
4028         e1000_init_manageability(adapter);
4029
4030         if (netif_running(netdev))
4031                 e1000e_up(adapter);
4032
4033         netif_device_attach(netdev);
4034
4035         /*
4036          * If the controller has AMT, do not set DRV_LOAD until the interface
4037          * is up.  For all other cases, let the f/w know that the h/w is now
4038          * under the control of the driver.
4039          */
4040         if (!(adapter->flags & FLAG_HAS_AMT) || !e1000e_check_mng_mode(&adapter->hw))
4041                 e1000_get_hw_control(adapter);
4042
4043         return 0;
4044 }
4045 #endif
4046
4047 static void e1000_shutdown(struct pci_dev *pdev)
4048 {
4049         e1000_suspend(pdev, PMSG_SUSPEND);
4050 }
4051
4052 #ifdef CONFIG_NET_POLL_CONTROLLER
4053 /*
4054  * Polling 'interrupt' - used by things like netconsole to send skbs
4055  * without having to re-enable interrupts. It's not called while
4056  * the interrupt routine is executing.
4057  */
4058 static void e1000_netpoll(struct net_device *netdev)
4059 {
4060         struct e1000_adapter *adapter = netdev_priv(netdev);
4061
4062         disable_irq(adapter->pdev->irq);
4063         e1000_intr(adapter->pdev->irq, netdev);
4064
4065         e1000_clean_tx_irq(adapter);
4066
4067         enable_irq(adapter->pdev->irq);
4068 }
4069 #endif
4070
4071 /**
4072  * e1000_io_error_detected - called when PCI error is detected
4073  * @pdev: Pointer to PCI device
4074  * @state: The current pci connection state
4075  *
4076  * This function is called after a PCI bus error affecting
4077  * this device has been detected.
4078  */
4079 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4080                                                 pci_channel_state_t state)
4081 {
4082         struct net_device *netdev = pci_get_drvdata(pdev);
4083         struct e1000_adapter *adapter = netdev_priv(netdev);
4084
4085         netif_device_detach(netdev);
4086
4087         if (netif_running(netdev))
4088                 e1000e_down(adapter);
4089         pci_disable_device(pdev);
4090
4091         /* Request a slot slot reset. */
4092         return PCI_ERS_RESULT_NEED_RESET;
4093 }
4094
4095 /**
4096  * e1000_io_slot_reset - called after the pci bus has been reset.
4097  * @pdev: Pointer to PCI device
4098  *
4099  * Restart the card from scratch, as if from a cold-boot. Implementation
4100  * resembles the first-half of the e1000_resume routine.
4101  */
4102 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4103 {
4104         struct net_device *netdev = pci_get_drvdata(pdev);
4105         struct e1000_adapter *adapter = netdev_priv(netdev);
4106         struct e1000_hw *hw = &adapter->hw;
4107
4108         e1000e_disable_l1aspm(pdev);
4109         if (pci_enable_device(pdev)) {
4110                 dev_err(&pdev->dev,
4111                         "Cannot re-enable PCI device after reset.\n");
4112                 return PCI_ERS_RESULT_DISCONNECT;
4113         }
4114         pci_set_master(pdev);
4115         pci_restore_state(pdev);
4116
4117         pci_enable_wake(pdev, PCI_D3hot, 0);
4118         pci_enable_wake(pdev, PCI_D3cold, 0);
4119
4120         e1000e_reset(adapter);
4121         ew32(WUS, ~0);
4122
4123         return PCI_ERS_RESULT_RECOVERED;
4124 }
4125
4126 /**
4127  * e1000_io_resume - called when traffic can start flowing again.
4128  * @pdev: Pointer to PCI device
4129  *
4130  * This callback is called when the error recovery driver tells us that
4131  * its OK to resume normal operation. Implementation resembles the
4132  * second-half of the e1000_resume routine.
4133  */
4134 static void e1000_io_resume(struct pci_dev *pdev)
4135 {
4136         struct net_device *netdev = pci_get_drvdata(pdev);
4137         struct e1000_adapter *adapter = netdev_priv(netdev);
4138
4139         e1000_init_manageability(adapter);
4140
4141         if (netif_running(netdev)) {
4142                 if (e1000e_up(adapter)) {
4143                         dev_err(&pdev->dev,
4144                                 "can't bring device back up after reset\n");
4145                         return;
4146                 }
4147         }
4148
4149         netif_device_attach(netdev);
4150
4151         /*
4152          * If the controller has AMT, do not set DRV_LOAD until the interface
4153          * is up.  For all other cases, let the f/w know that the h/w is now
4154          * under the control of the driver.
4155          */
4156         if (!(adapter->flags & FLAG_HAS_AMT) ||
4157             !e1000e_check_mng_mode(&adapter->hw))
4158                 e1000_get_hw_control(adapter);
4159
4160 }
4161
4162 static void e1000_print_device_info(struct e1000_adapter *adapter)
4163 {
4164         struct e1000_hw *hw = &adapter->hw;
4165         struct net_device *netdev = adapter->netdev;
4166         u32 pba_num;
4167
4168         /* print bus type/speed/width info */
4169         ndev_info(netdev, "(PCI Express:2.5GB/s:%s) "
4170                   "%02x:%02x:%02x:%02x:%02x:%02x\n",
4171                   /* bus width */
4172                  ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4173                   "Width x1"),
4174                   /* MAC address */
4175                   netdev->dev_addr[0], netdev->dev_addr[1],
4176                   netdev->dev_addr[2], netdev->dev_addr[3],
4177                   netdev->dev_addr[4], netdev->dev_addr[5]);
4178         ndev_info(netdev, "Intel(R) PRO/%s Network Connection\n",
4179                   (hw->phy.type == e1000_phy_ife)
4180                    ? "10/100" : "1000");
4181         e1000e_read_pba_num(hw, &pba_num);
4182         ndev_info(netdev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4183                   hw->mac.type, hw->phy.type,
4184                   (pba_num >> 8), (pba_num & 0xff));
4185 }
4186
4187 /**
4188  * e1000_probe - Device Initialization Routine
4189  * @pdev: PCI device information struct
4190  * @ent: entry in e1000_pci_tbl
4191  *
4192  * Returns 0 on success, negative on failure
4193  *
4194  * e1000_probe initializes an adapter identified by a pci_dev structure.
4195  * The OS initialization, configuring of the adapter private structure,
4196  * and a hardware reset occur.
4197  **/
4198 static int __devinit e1000_probe(struct pci_dev *pdev,
4199                                  const struct pci_device_id *ent)
4200 {
4201         struct net_device *netdev;
4202         struct e1000_adapter *adapter;
4203         struct e1000_hw *hw;
4204         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4205         resource_size_t mmio_start, mmio_len;
4206         resource_size_t flash_start, flash_len;
4207
4208         static int cards_found;
4209         int i, err, pci_using_dac;
4210         u16 eeprom_data = 0;
4211         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4212
4213         e1000e_disable_l1aspm(pdev);
4214         err = pci_enable_device(pdev);
4215         if (err)
4216                 return err;
4217
4218         pci_using_dac = 0;
4219         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4220         if (!err) {
4221                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4222                 if (!err)
4223                         pci_using_dac = 1;
4224         } else {
4225                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4226                 if (err) {
4227                         err = pci_set_consistent_dma_mask(pdev,
4228                                                           DMA_32BIT_MASK);
4229                         if (err) {
4230                                 dev_err(&pdev->dev, "No usable DMA "
4231                                         "configuration, aborting\n");
4232                                 goto err_dma;
4233                         }
4234                 }
4235         }
4236
4237         err = pci_request_regions(pdev, e1000e_driver_name);
4238         if (err)
4239                 goto err_pci_reg;
4240
4241         pci_set_master(pdev);
4242         pci_save_state(pdev);
4243
4244         err = -ENOMEM;
4245         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4246         if (!netdev)
4247                 goto err_alloc_etherdev;
4248
4249         SET_NETDEV_DEV(netdev, &pdev->dev);
4250
4251         pci_set_drvdata(pdev, netdev);
4252         adapter = netdev_priv(netdev);
4253         hw = &adapter->hw;
4254         adapter->netdev = netdev;
4255         adapter->pdev = pdev;
4256         adapter->ei = ei;
4257         adapter->pba = ei->pba;
4258         adapter->flags = ei->flags;
4259         adapter->hw.adapter = adapter;
4260         adapter->hw.mac.type = ei->mac;
4261         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4262
4263         mmio_start = pci_resource_start(pdev, 0);
4264         mmio_len = pci_resource_len(pdev, 0);
4265
4266         err = -EIO;
4267         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4268         if (!adapter->hw.hw_addr)
4269                 goto err_ioremap;
4270
4271         if ((adapter->flags & FLAG_HAS_FLASH) &&
4272             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4273                 flash_start = pci_resource_start(pdev, 1);
4274                 flash_len = pci_resource_len(pdev, 1);
4275                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4276                 if (!adapter->hw.flash_address)
4277                         goto err_flashmap;
4278         }
4279
4280         /* construct the net_device struct */
4281         netdev->open                    = &e1000_open;
4282         netdev->stop                    = &e1000_close;
4283         netdev->hard_start_xmit         = &e1000_xmit_frame;
4284         netdev->get_stats               = &e1000_get_stats;
4285         netdev->set_multicast_list      = &e1000_set_multi;
4286         netdev->set_mac_address         = &e1000_set_mac;
4287         netdev->change_mtu              = &e1000_change_mtu;
4288         netdev->do_ioctl                = &e1000_ioctl;
4289         e1000e_set_ethtool_ops(netdev);
4290         netdev->tx_timeout              = &e1000_tx_timeout;
4291         netdev->watchdog_timeo          = 5 * HZ;
4292         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4293         netdev->vlan_rx_register        = e1000_vlan_rx_register;
4294         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
4295         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
4296 #ifdef CONFIG_NET_POLL_CONTROLLER
4297         netdev->poll_controller         = e1000_netpoll;
4298 #endif
4299         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4300
4301         netdev->mem_start = mmio_start;
4302         netdev->mem_end = mmio_start + mmio_len;
4303
4304         adapter->bd_number = cards_found++;
4305
4306         /* setup adapter struct */
4307         err = e1000_sw_init(adapter);
4308         if (err)
4309                 goto err_sw_init;
4310
4311         err = -EIO;
4312
4313         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4314         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4315         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4316
4317         err = ei->get_variants(adapter);
4318         if (err)
4319                 goto err_hw_init;
4320
4321         hw->mac.ops.get_bus_info(&adapter->hw);
4322
4323         adapter->hw.phy.autoneg_wait_to_complete = 0;
4324
4325         /* Copper options */
4326         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4327                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4328                 adapter->hw.phy.disable_polarity_correction = 0;
4329                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4330         }
4331
4332         if (e1000_check_reset_block(&adapter->hw))
4333                 ndev_info(netdev,
4334                           "PHY reset is blocked due to SOL/IDER session.\n");
4335
4336         netdev->features = NETIF_F_SG |
4337                            NETIF_F_HW_CSUM |
4338                            NETIF_F_HW_VLAN_TX |
4339                            NETIF_F_HW_VLAN_RX;
4340
4341         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4342                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4343
4344         netdev->features |= NETIF_F_TSO;
4345         netdev->features |= NETIF_F_TSO6;
4346
4347         if (pci_using_dac)
4348                 netdev->features |= NETIF_F_HIGHDMA;
4349
4350         /*
4351          * We should not be using LLTX anymore, but we are still Tx faster with
4352          * it.
4353          */
4354         netdev->features |= NETIF_F_LLTX;
4355
4356         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4357                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4358
4359         /*
4360          * before reading the NVM, reset the controller to
4361          * put the device in a known good starting state
4362          */
4363         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4364
4365         /*
4366          * systems with ASPM and others may see the checksum fail on the first
4367          * attempt. Let's give it a few tries
4368          */
4369         for (i = 0;; i++) {
4370                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4371                         break;
4372                 if (i == 2) {
4373                         ndev_err(netdev, "The NVM Checksum Is Not Valid\n");
4374                         err = -EIO;
4375                         goto err_eeprom;
4376                 }
4377         }
4378
4379         /* copy the MAC address out of the NVM */
4380         if (e1000e_read_mac_addr(&adapter->hw))
4381                 ndev_err(netdev, "NVM Read Error while reading MAC address\n");
4382
4383         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4384         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4385
4386         if (!is_valid_ether_addr(netdev->perm_addr)) {
4387                 ndev_err(netdev, "Invalid MAC Address: "
4388                          "%02x:%02x:%02x:%02x:%02x:%02x\n",
4389                          netdev->perm_addr[0], netdev->perm_addr[1],
4390                          netdev->perm_addr[2], netdev->perm_addr[3],
4391                          netdev->perm_addr[4], netdev->perm_addr[5]);
4392                 err = -EIO;
4393                 goto err_eeprom;
4394         }
4395
4396         init_timer(&adapter->watchdog_timer);
4397         adapter->watchdog_timer.function = &e1000_watchdog;
4398         adapter->watchdog_timer.data = (unsigned long) adapter;
4399
4400         init_timer(&adapter->phy_info_timer);
4401         adapter->phy_info_timer.function = &e1000_update_phy_info;
4402         adapter->phy_info_timer.data = (unsigned long) adapter;
4403
4404         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4405         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4406
4407         e1000e_check_options(adapter);
4408
4409         /* Initialize link parameters. User can change them with ethtool */
4410         adapter->hw.mac.autoneg = 1;
4411         adapter->fc_autoneg = 1;
4412         adapter->hw.fc.original_type = e1000_fc_default;
4413         adapter->hw.fc.type = e1000_fc_default;
4414         adapter->hw.phy.autoneg_advertised = 0x2f;
4415
4416         /* ring size defaults */
4417         adapter->rx_ring->count = 256;
4418         adapter->tx_ring->count = 256;
4419
4420         /*
4421          * Initial Wake on LAN setting - If APM wake is enabled in
4422          * the EEPROM, enable the ACPI Magic Packet filter
4423          */
4424         if (adapter->flags & FLAG_APME_IN_WUC) {
4425                 /* APME bit in EEPROM is mapped to WUC.APME */
4426                 eeprom_data = er32(WUC);
4427                 eeprom_apme_mask = E1000_WUC_APME;
4428         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4429                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4430                     (adapter->hw.bus.func == 1))
4431                         e1000_read_nvm(&adapter->hw,
4432                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4433                 else
4434                         e1000_read_nvm(&adapter->hw,
4435                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4436         }
4437
4438         /* fetch WoL from EEPROM */
4439         if (eeprom_data & eeprom_apme_mask)
4440                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4441
4442         /*
4443          * now that we have the eeprom settings, apply the special cases
4444          * where the eeprom may be wrong or the board simply won't support
4445          * wake on lan on a particular port
4446          */
4447         if (!(adapter->flags & FLAG_HAS_WOL))
4448                 adapter->eeprom_wol = 0;
4449
4450         /* initialize the wol settings based on the eeprom settings */
4451         adapter->wol = adapter->eeprom_wol;
4452
4453         /* reset the hardware with the new settings */
4454         e1000e_reset(adapter);
4455
4456         /*
4457          * If the controller has AMT, do not set DRV_LOAD until the interface
4458          * is up.  For all other cases, let the f/w know that the h/w is now
4459          * under the control of the driver.
4460          */
4461         if (!(adapter->flags & FLAG_HAS_AMT) ||
4462             !e1000e_check_mng_mode(&adapter->hw))
4463                 e1000_get_hw_control(adapter);
4464
4465         /* tell the stack to leave us alone until e1000_open() is called */
4466         netif_carrier_off(netdev);
4467         netif_stop_queue(netdev);
4468
4469         strcpy(netdev->name, "eth%d");
4470         err = register_netdev(netdev);
4471         if (err)
4472                 goto err_register;
4473
4474         e1000_print_device_info(adapter);
4475
4476         return 0;
4477
4478 err_register:
4479 err_hw_init:
4480         e1000_release_hw_control(adapter);
4481 err_eeprom:
4482         if (!e1000_check_reset_block(&adapter->hw))
4483                 e1000_phy_hw_reset(&adapter->hw);
4484
4485         if (adapter->hw.flash_address)
4486                 iounmap(adapter->hw.flash_address);
4487
4488 err_flashmap:
4489         kfree(adapter->tx_ring);
4490         kfree(adapter->rx_ring);
4491 err_sw_init:
4492         iounmap(adapter->hw.hw_addr);
4493 err_ioremap:
4494         free_netdev(netdev);
4495 err_alloc_etherdev:
4496         pci_release_regions(pdev);
4497 err_pci_reg:
4498 err_dma:
4499         pci_disable_device(pdev);
4500         return err;
4501 }
4502
4503 /**
4504  * e1000_remove - Device Removal Routine
4505  * @pdev: PCI device information struct
4506  *
4507  * e1000_remove is called by the PCI subsystem to alert the driver
4508  * that it should release a PCI device.  The could be caused by a
4509  * Hot-Plug event, or because the driver is going to be removed from
4510  * memory.
4511  **/
4512 static void __devexit e1000_remove(struct pci_dev *pdev)
4513 {
4514         struct net_device *netdev = pci_get_drvdata(pdev);
4515         struct e1000_adapter *adapter = netdev_priv(netdev);
4516
4517         /*
4518          * flush_scheduled work may reschedule our watchdog task, so
4519          * explicitly disable watchdog tasks from being rescheduled
4520          */
4521         set_bit(__E1000_DOWN, &adapter->state);
4522         del_timer_sync(&adapter->watchdog_timer);
4523         del_timer_sync(&adapter->phy_info_timer);
4524
4525         flush_scheduled_work();
4526
4527         /*
4528          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4529          * would have already happened in close and is redundant.
4530          */
4531         e1000_release_hw_control(adapter);
4532
4533         unregister_netdev(netdev);
4534
4535         if (!e1000_check_reset_block(&adapter->hw))
4536                 e1000_phy_hw_reset(&adapter->hw);
4537
4538         kfree(adapter->tx_ring);
4539         kfree(adapter->rx_ring);
4540
4541         iounmap(adapter->hw.hw_addr);
4542         if (adapter->hw.flash_address)
4543                 iounmap(adapter->hw.flash_address);
4544         pci_release_regions(pdev);
4545
4546         free_netdev(netdev);
4547
4548         pci_disable_device(pdev);
4549 }
4550
4551 /* PCI Error Recovery (ERS) */
4552 static struct pci_error_handlers e1000_err_handler = {
4553         .error_detected = e1000_io_error_detected,
4554         .slot_reset = e1000_io_slot_reset,
4555         .resume = e1000_io_resume,
4556 };
4557
4558 static struct pci_device_id e1000_pci_tbl[] = {
4559         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4560         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4561         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4562         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4563         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4564         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
4565         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
4566         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
4567         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
4568
4569         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4570         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4571         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4572         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4573
4574         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4575         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4576         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4577
4578         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4579           board_80003es2lan },
4580         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4581           board_80003es2lan },
4582         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4583           board_80003es2lan },
4584         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4585           board_80003es2lan },
4586
4587         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4588         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4589         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4590         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4591         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4592         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4593         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
4594
4595         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4596         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4597         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4598         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4599         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4600         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
4601         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
4602         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
4603
4604         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
4605         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
4606         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
4607
4608         { }     /* terminate list */
4609 };
4610 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4611
4612 /* PCI Device API Driver */
4613 static struct pci_driver e1000_driver = {
4614         .name     = e1000e_driver_name,
4615         .id_table = e1000_pci_tbl,
4616         .probe    = e1000_probe,
4617         .remove   = __devexit_p(e1000_remove),
4618 #ifdef CONFIG_PM
4619         /* Power Management Hooks */
4620         .suspend  = e1000_suspend,
4621         .resume   = e1000_resume,
4622 #endif
4623         .shutdown = e1000_shutdown,
4624         .err_handler = &e1000_err_handler
4625 };
4626
4627 /**
4628  * e1000_init_module - Driver Registration Routine
4629  *
4630  * e1000_init_module is the first routine called when the driver is
4631  * loaded. All it does is register with the PCI subsystem.
4632  **/
4633 static int __init e1000_init_module(void)
4634 {
4635         int ret;
4636         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4637                e1000e_driver_name, e1000e_driver_version);
4638         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
4639                e1000e_driver_name);
4640         ret = pci_register_driver(&e1000_driver);
4641         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
4642                                PM_QOS_DEFAULT_VALUE);
4643
4644         return ret;
4645 }
4646 module_init(e1000_init_module);
4647
4648 /**
4649  * e1000_exit_module - Driver Exit Cleanup Routine
4650  *
4651  * e1000_exit_module is called just before the driver is removed
4652  * from memory.
4653  **/
4654 static void __exit e1000_exit_module(void)
4655 {
4656         pci_unregister_driver(&e1000_driver);
4657         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
4658 }
4659 module_exit(e1000_exit_module);
4660
4661
4662 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4663 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4664 MODULE_LICENSE("GPL");
4665 MODULE_VERSION(DRV_VERSION);
4666
4667 /* e1000_main.c */